Professional Documents
Culture Documents
OF
VOLUME 11
\ V jFm
'
/ /
NUMBER 3
CONTENTS
PART I - ADVANCED
Some Generalizations Suggested by Gould1 s
Systematic Treatment of Certain Binomial Identities .
. . . . . .
F. W. Light, Jr.
. . . / ? . . Buschman 247
241
. . . . . . .
. . . . . . . . .
. Haruo Hosoya
255
Marjorie Bicknell
266
David A, Klarner
267
271
Exponential Generating Functions for Fibonacci Identities , C. A. Church and Marjorie Bicknell
Notes on Binomial Coefficients: IV Proof of a Conjecture
of Gould on the GCD ! s of Two Triples of Binomial Coefficients
275
L e t t e r s to the Editors .
284
285
R. G. Buschman
Multiple Reflections.
295
. . . .
Dean R. Hickerson
302
307
312
PART II - ELEMENTARY
An Inequality in a Certain Diophantine Equation.
A P r i m e r for the
Fibonacci Numbers: P a r t XII . .
OCTOBER
0. A Butter
315
332
333
1973
317
CO-EDITORS
V. E. Hoggatt, J r .
Marjorie Bicknell
EDITORIAL BOARD
A. P . Hillman
David A. Klarner
Donald E. Knuth
C. T. Long
M. N. S. Swamy
D. E. Thoro
IL L. Alder
John L. Brown, J r .
Brother A. Brousseau
Paul F. Byrd
L. Carlitz
H. W. Eves
H. W. Gould
D. A. Und
J a m e s Maxwell
Sister M. DeSales McNabb
Gerald Preston
D. W. Robinson
Azriel Roselfeld
Lloyd Walker
Charles H. Wall
The California Mathematics Council
All subscription correspondence should be addressed to Brother Alfred Brousseau, St. Mary f s
College, California 94575. All checks ($8.00 per year) should be made out to the Fibonacci
Association or the Fibonacci Quarterly.
Drawings should be made the same size as they will appear in the Quarterly, and
(1)
f(n) = (1 - u)-'(l + uH = E A
n=0
In addition, Professor H. W. Gould of West Virginia University, in a recently published paper [ 2 ] , has indicated several additional identities for the A
coefficients.
We now introduce the generalized numbers A (x) defined by the generating function
(2)
g(u,x) = (1 - u T ^ l + u) X = ] T A n ( x ) u n ,
n=0
which is valid for all real or complex x; from this, the following relations a r e evident:
(3)
f ( u ) . = g(u,-.J)
(4)
(5)
A n (x)
= A (- \)
k=0
(1)=H> k ( 2 k k )
225
226
[Oct
The purpose of this paper i s to present some properties of the coefficients A (x).
Gould [2] has demonstrated that most of the identities shown in [ l ] are old r e s u l t s , citing
numerous references to substantiate this claim.
the w r i t e r , Gould has indicated that the coefficients A (x) have been studied extensively by
previous mathematicians.
(8)
Gould, who as well as being a mathematician of the highest order, i s an expert in the field of
information retrieval, was impressed by the apparent novelty of the relation in (8), and his
closing r e m a r k s in [2 ] stimulated a search for a suitable generalization of (8). This search
was initiated by the writer in an effort to find a single-sum expression for the coefficient
A2 (x). In this respect, he has failed.
eralization of (8) by empirical methods, and this is expressed in the following elegant form:
(9)
(k=0
I k=0
'I
k=0
2"
k=0
= 0 .
F o r x = - \ , it b e -
(ID
An(X) - (-i> S
/ w
\
k
(-D (j)(;\-i) ;
k=0
derived by expressing g(u,x) in the form (1 - u)~ ~ x ( l - u 2 ) , and obtaining the convolute.
1973]
227
w-i-u(i)i:^(i)Hi '
(12)
k=0
'
~ l |/-,l ^
+ r
2u
- j\
viously stated in variant form a s formula (22) in [ l ] . NOTE: Identity (12) has been submitted to Advanced Problems Editor a s a proposed problem.
(*)
n
0k
A. (x)
(x) == f r \ y
) W
, H
2J x.
x - k
(13)
derived by obtaining the convolute of the function g(u,x) expressed in the form
(1 + u)
For x = - 1 , this becomes formula (13) in [2].
U2 A-2X-1e -tQ2 t = ^
^ JI t L
>z
00
u2n ^ ( - l ) k u 2 k - 2 x
x
k
i(2k-2x)
l n
n=0
- k=0
n=0 " ""
x
dt =ZJ-^ZSL,
(14)
(-l)nAn(x)u2n-2X
" 2^
7~7
(0
2n - \
i-i
(k)
(- > = ( j )k=0
S ^> fj: \ ^ ^ - ^
(16)
,z^
^ k w
Relations (15) and (16) a r e obtained from (12) and (13) by inversion.
F o r x = - j , (15) and
(16) become (15) and (16), respectively, in [ 2 ] , which Gould obtained by the same method.
(17)
W (k)
n
1
^
{
i
-i)
(^)"
}=
L
2 {1 ++ ((-1)" C ] > = V V - f A* 2 k(x) x - 2k
1
k=0
ol
228
i (
as)
L 2 J (2kn+ i )
/ \-i)
x n
*
A
2k+1(x) x _ -k_ 1 .
1)
a u - c - D - m > = => : (^
1
[Oct.
2k
Relations (17) and (18) are obtained by respectively adding and subtracting (15) and (16).
When x = - -|,
(l-u^)
/ -i\ n 2n-2x
( 1} U
/ t" ^ !! - t * ) d t =
(19)
/x
" ( - l ) n + 1 A (x)
^2n-2x
" ^(x-n)f
(20)
J?t-2x-l
/ 7T7
dt =
1)nu2n-2x
n=0
(2-uV
( - l ) n + 1 A (x)
2-isnrsr = 2J
(x.n)M
n=0
u 2n-2x
^ - ^
v
I n J
1
(21)
E
(
n=0
/ .x - l\ ,
lV n
'
k2k-2x
2 n V / x \ (-l)KuZK"Zx
2 J
k=0
(2k- 2x)
Z,(kX P W ^ x 7 - S (x n - k ) ^
k=0
_
v^ a
2,
n=0
;2n_2x
OO
Y ^ / x \ ( - l ) k u 2 k ~ 2 x V-* / - x - l \ ,
' 0
OO
(2t2 - l ) n d t
)(-D
f 2x " 1
'
. n 2n-2x \~"*
lX
( L)
U
n=k
n=0
Ls
-vii-k :
U
'
(2k - 2xT
k=0
We will return to the above expression, but first we will direct our efforts toward finding an
expression for 6 , as defined above. If we differentiate the integral expression for y and
its s e r i e s form:
OO
2x
2 X
" (l - u )
1973]
/.
(1 - u )y' = u " ^
229
+ 2u(x + l ) y ;
converting this differential equation into s e r i e s form by means of the foregoing relations, we
obtain a recursion: (n + 1 - x>^ n + 1 = (n + D^ n (n = 0, 1, 2, ); 1= -2x6Q
(x ? 0).
By an
>n = ( - l ) n + 1 / 2(x
this proves the first identity of (19). We may convert such form as follows, by use of (16):
(-1) A k (x)
-l/n\
2n-2x
^- W
0
k=0
y =
^5"
(-1)
Z-^
k=0
OO
(-1) Afe(x)
A (x)2
( x
^/x\
"
k )
-n-k-l/n+k\
k u
( )
2k-2x/n
! r
2n+2k-2x
k-l
d-iu )
which reduces to (19). If we return to the double summation expression for y which we first
obtained, we a r r i v e at the interesting identity:
= yMM\ n - k j
(22)
:-nx
= "E
.(!)
(X
xV
lf
k = ( )
Relation (20) is similarly obtained from (15) being substituted in the first identity of (20), which
is readily obtained from the integral expression by direct integration of a geometric s e r i e s .
Relation (21) is derived from (12), and may be verified by expansion of t h e integrand in
(21), t e r m - b y - t e r m integration and comparison with (12).
A (x)
(x
n*
)
_
(iu)
n=0 _
/__ " _\ l/ nx \I f f
2
^p
2n
J (u)u n
y
*L__
= n=0
^ ( x - n)n!
230
function of n
[Oct.
2n-2x
J r ^ a - t ^ d t = (-!)*(*)!
2n -
(24)
2x
n=0
Relation (24) is obtained by expansion of the integrand and t e r m - b y - t e r m integration of
the result.
By the substitution of x = - i
(18), (19), (21) (in variant form), (28) and (35) in [2].
By the substitutions u = iv, t = is in (19), (20) and (24) (and reconverting back to the
dummy variables u and t), we derive the following:
(25)
v-X-
(1 + u22 )
-1 f t - ^ t t + tffdt = y;2n-2x
0J
n=0
(2n
2X)
"
A (x)
ft
2n-2x
^ ( n - x ^ M (2 + u*) n + 1
n=0
U
(26)
/0
-2x-l
^ *
J l , 2n-2x
^
~2X=^l
^n=0
J!,
An(x)
u 2n-2x
x
n=0 ( n - x )I/ n \J
2n-2x
(27)
2x
2 - u
v2,
- * - = *
2 - t2
(28)
(1 - u ^ -
f r *-
1 (1
-t2) x
1 + t2
A / \ 2n-2x
dt = y_ _; (-Dn As.(x) u
n=Q
(2n - 2x)
15
1973]
(29)
2 x 1 ( 1 + t2)
u*)-*- f r -
1- -" t-2
/0
oo
231
A /~\ 2 n - 2 x
dt = y
((2n
2n
n =0
2x)
ft'
Another genus of relations is obtained by considering variations inform of the basic definition
of g(u,x) in (2), or related functions.
F o r example, since
(l + u ) x + r ( l
u ) - x + s = (1 + u ) r + s
we a r r i v e at
This is simply a special case of the Vandermonde convolution theorem; its chief point
of interest here is the invariance of (30) with respect to x.
Setting r = 0 and
s = -1,
as
k=0
(32)
k=0
A(-r
- Ax^v
( r + s> .
fcV* 4 s)
0/
'
Again, the interesting point in (32) is the invariance with respect to x of the right member.
When r = 0, s = x, we obtain the
th expression in (5) for A (x). By setting x = 0 and s
x in (32), we obtain the recursion:
Vx
(33)
+ r)
=2.(kJAn-k(x)
k=0
Another interesting identity displaying invariance on x is obtained by considering the convolution of (l + u ^ d - u ) "
(l + u ) ~ x + s ( l - u ) " 1 = (l + u ) r + s ( l - u ) ~ 2 .
232
[Oct.
n
(34)
^ Ak(x
k=0
+ r)A
n-k
""x
s)
(n +
1)A
n(r
s)
(r +
s)A
n-l(r
+ s
"
X)
'
s = 0, we obtain:
^2 A k ( x ) A n-k ( " x)
(35)
= n + 1
'
k=0
For r = 0, s = - 1 , (34) yields:
(36)
Yl
k ( x ) A n-k ( - x - 1) = 1 +
[^]
k=0
By considering the sum
(l + u ) x + k
1 - u
(1 + u) x ^ (1 + u ) r - 1
1 - u
u
k=0
k=0
A (x + 1) = A (x) + A , (x) .
n
nv
n-1
We may also derive (38) by letting r = 1 in (33). As should otherwise be evident, this is
the same recursion satisfied by the binomial coefficients, i. e. , if j
I is substituted for
A v(x).
n
The list of identities in (10)-(38) is by no means exhaustive, and indeed it should have
by now become evident to the r e a d e r that the variety of derivable identities stemming from
the basic definition in (2) is virtually unlimited. As previously intimated, Gould [2] has observed that far more general results a r e available in the existing l i t e r a t u r e , and it is p r i m a r i l y for this reason that (10)-(38) have been offered with a minimum of explanation.
The
real purpose of this paper is to give a proof of (9), and the other identities have been pre sen ted
1973]
233
direct proof, and this is left as a project for the interested reader.
We begin by adopting the following definitions, in the interest of simplicity of expression:
(39)
Cn = A n ( x - | ) ;
Cn=An(-x-^)
(40)
Qn = c n cr n
(41)
Rn = Qn - Q n _ 1
^-(':<) ^-M
(42)
(43)
'
(44)
n
q = Kt
= J J
n n
= \ - *
Some useful relations a r e indicated below, which a r e evident from the definitions given
in (39)-(44):
(45)
v
'
C = C 1 + J ;
n
n-1
n
(46)
n
n
C1-1),
\
I^
-x J
\({-n)>.
(47)
K = { n
i;
^ n .
I n-1 '
2
I K ,
( T = C " 1 + J
n
n-1
n
J -r ''>
n
("-l>
n2
n-1
1
Q ,
same recursion, with the same initial conditions, is satisfied by the expression in the right
m e m b e r of (9). The following development makes free use of the relations and definitions in
(39)-(47):
R
= C C " - C
C * , = C C " - ( C
- J )7 ( C " _ J ) ,
n n
n - 11 n - 1
n n
n
n
n
n
or
v(48)
R = J C " + J C - K .
n
n n
n n
n
'
If we increase the subscript in (48) by unity, multiply by (n + 1), and apply (45)-(47),
we obtain:
(n
l ) R n + 1 = (n
l ) J n + 1 ( C n + J n + 1 ) + (n
1 ) 3 ^
+ J n + 1 ) - (n + D K n + 1 .
or
(49)
+ (n + 1)K a . 1 .
n+1
If we decrease the subscript in (48) by unity, and again use relations (45)-(47), we
obtain:
234
[Oct.
- = C" - J - + C n J , - K
, = (C -1)1
2
) J
n-1
n-1 n-1
n-1 n-1
n-1
n
n 1
, 1
I n
\x + L - n/
a Ja)
+ (c.
llJ^LzJL,
n
Tn
(q + n f a - D )
n
= (
n-1
.x
q+n
^"^'
i n)(c _ J )j
2
n
n n
+ v(x + SL _n)(C - J )J - nK ,
2
n
n n
n'
+
or
(50)
i - i
nn
*
nn
2
(i* -')-! =
+ (n - 1)K
n
If we now multiply (48) throughout by 2n and add this result to the sum of (49) and (50),
we obtain the following recursion:
(51)
(n
+ 1)R ^ + 2nR + ( ^ + n - l ) R
., = v(n + 1)(K n - K )
v
n+1
n
In
I n-1
^ n+1
n
If, in (51), we substitute for R
the expression Q - Q
(n
l ) Q n + 1 + (n - l)Q n
^ _n _ XJ Q ^
. ^
+ n
. ^
Q^
This, then, is the recursion which we now seek to demonstrate is also satisfied by the
expression in the right member of (9).
We begin by introducing some additional definitions, again for the sake of brevity:
n
(53)
^ P = J >J J .
n
n Z-> n-k
k=0
(54)
- \: "
t
- 2 ~
_
P
;
, '
k=0
_ ^
x + i + n
= J > J .
n
n Z^r n-k
,
t
i.
"
dt ;
h = w(u, -x) .
The statement of identity (9) may then be condensed to the simple form:
(55)
Q n ^ P n
+ Pn*-
Since
n=0 ^
197
3]
235
and
u
-x~J
ff^--z^
n=0
k=0
n=0
-x+j-+n
x
" 2 "
l \
/ x -
- k
n=Q
k=()
x - - - k
Comparing the latter expression with the first definition in (53), we have:
P
(56)
u"x+^+n
h = ^2
n=0
P
;
similarly,
h = ^
(x - \ - n)J n
n=()
+x+
T+n
(-x - \ - n)J n
h' =
2i
n+1
= - P 0 i T X - ^ + Y\
n=-l
T
x
h = (x + ^ ) P 0 u - -
o/o
3/2
(58)
= (x + ^ ) P 0 u
_x
X _ 3/2
3/2
(x - i - n)J
*
n
n=0
(n + 2 ) P n + 9 u _ x + - 2 - + n
22
"
+ V
i
V - (n
+ P l U - "-2 + J^
~
n=0
Xx
+ 1)(n +
2)P
n+2U
(x - -I - n ) J
2
n
-x+i+n
2
)-X_3/2 /
0
\4
dt
or:
(1 - u 2 )h' + (x + ^)(1 - u)h + (1 + u)" X + ^ u - * ^
(59)
By a second differentiation,
2
/ x -i- -
(\ X+^I
= 0 .
x -x -= i
-i)
+TTr4!(1
u)"x+^u"x~2
236
[Oct.
By means of the s e r i e s form for h in (56), for h ! in (57), and the identity:
n )u
(i + u) ^u ^ = 2 ^ 1
n=0
'
n=0
we may convert (59) entirely into s e r i e s form, with certain manipulations based on p r o p e r ties of the binomial coefficients, designed to maintain the exponent of u in the various expressions the same (-x + \ + n, in our development), and to contain the factor J
in the
^
n
denominator of each expression, which may subsequently be cancelled. If we do so, we obtain the following:
" -(n + 1)P
, u-*^"*1
;
V^ v
n+1
n=0
n=0
(*^-^n
^x " ? -
n)J
J ^
V*
(x
VA
+ -L - ll)n ) P
u-***4*
2
n-1
n=l
n=0
= 0 .
If we now equate the coefficients of similar powers of u in the above expression and
simplify by multiplying throughout by (x - |- - n) J , we obtain:
PA = q + 2 x ;
-(n + l ) P n + 1 - (x - \ - n ) P n ^ + (x + \)V^
K
= (x - fr)(x - J. - n) j i L - ,
or
(61)
(n + l)(P ^ - P , ) - (x + i ) ( P - P
n) = < - 4 - T + x - i > K
7
n+1
n-1
2/v n
n-1
)n + 1
21 n
P 0 = 1,
P i = q + 2x
(n = l , 2 , - - - )
By a s i m i l a r , though more complicated manipulation of the s e r i e s in (56)-(58), we may exp r e s s (60) in s e r i e s form, yielding another recursion for the
omitted h e r e , since it is somewhat lengthy.
P's.
The development is
(n + D 2 r
i
^ n+1
x
n
( x(n + 2 ) + q + n2 + - | n )
( x n - q - n
"t -
with
* ^
Pi = q + 2x,
n-2
P 0 = 1,
'
(n = 2, 3, 4, )
P 2 = (^q - l ) z
+ X
Q .
- i n )
1973]
237
By multiplying the latter expression throughout by n(x - I - n), simplifying the result, and
shifting the t e r m s around, as the r e a d e r may verify, the following form of the recursion is
obtained:
(62)
"n(n
1)2(P
n+l -
+ n(
n-i>
+ n U
* ( P n "" P n-2>
_ 1)2 +
xq
V = P - P - ,
n
n
n-1
V
+ V
= P
- P
v
n
n+1
*n+l
n-1
(n
+ 1)(V ... - K , - ) = (x - J- - n)(V + K )
v
n+1
n+1
2
n
n
(n +
1)K
n+l "
nK
+ V n ) + (-Jq - n(n + ^ ) ) V n - 1
(65)
+ n(n + 2 ) x V n - ( q + n ( n - l ) ) x V _
= 0 .
(66)
xVn = -xKn
(n
i)(Vn
K n > + (n
l)(Vn+1 - Kn+1)
If we substitute the expression in (66) and the corresponding expression with the subscript
reduced by unity in (65), again use (47) in variant form, and simplify, (65) is transformed to
the following form:
(67)
(n + l ) V n + 1 + 2nV n + ( j + n - 1 j V n _ 1 = (n + l){K n + 1 - \ )
+ 2xK n .
The r e a d e r may verify the simplification to the above form, using the indicated p r o c e dure
(n + 1)P n+1
,- + (n - 1)P n + i[ ni - n ~ l J IP n - 1n - [ SI Ln +
(68)
Pnfs,
f
in accordance with
s:
- l \ Ip n-2
0
= fo + D(K n + 1 - K ) +-2xK .
n+1
n
n
238
[ 0 ct.
+ (n - 1)P
A1
n+1
(69)
+ | S L - n - l | P
n - | 2 +
I n
I n-1
In
n - l | P
/ n-
fc + wSfru-V- 5 "*
If we add (68) and (69) and divide by 2, we obtain the following recursion involving i (P + P ),
&. n
n
(n l ) , i ( P n + 1 + P^)
+ (n- l)i(Pn
the t e r m +s involving
x cancelling:
Pn)
^ -
2 +
Pn_2)
l ^ P ^
P ^
= (n + l ) ( K n + 1 - K n )
(70)
Comparing (52) with (70), we see that i (P '+P" ) satisfies the same recursion as Q . We
need to demonstrate only that Q = | ( P
P 2 = ( i q - l ) 2 + qx .
Pi = q + 2x,
Therefore,
i ( P 0 + P0) = 1 ;
^ ( P i + P i ) = ^(q + 2x + q - 2x) = q ;
4 (P 2 + P 2 )
= -*{(|q - D + q x + ( i q - l) 2 - qx } = (^q - l) 2 .
We may verify that
C0 = 1,
Ci = i + x,
C 2 = i x 2 + 7/8 = 1 - i q
Then
Qi = ( i + x)( l - x) = q,
Q2 = (1 - i q ) 2 .
How-
ever, certain results were obtained which suggest a r e a s of investigation for the interested
reader.
T
n
= A2 (x) - A2 n (x) .
nv '
n-lv
n^ - V l
()
We begin by
1973]
Tn ={An(x) - V j l x j f J A ^ + A ^ W l
239
or
(72)
T
n
Therefore,
11
k=i
= ( X J A (x + 1)
In I n
\
/
k=i x
k=i^
which yields:
(73)
A* W = > ,( J ]Afc(x + 1)
k=0
'
(74)
A^(x)
k=0
for A (x + 1) in:
n
(n + DA n + 1 (x + 1) - (x + 2)A n (x + 1) + (x + 1 - n J A ^ x + 1) = 0 ,
which is simply (10) with x + 1 replacing x.
(75)
T fs
= 0.
recursion:
(rjn\
J4U
(77)
ucl
"
that, in such c a s e , Q
= A
x = 0,
but if we observe
+ 1)(K
/v
x1 - K )
n+1
n
which
reduces to (77).)
In passing, we leave the reader with one possible form of expression for A2 (x), which
is suggested below by indicating the first few t e r m s
The writer
failed to see any pattern in the t e r m s of (78), but that is not to say that one does not exist.
The w r i t e r gratefully acknowledges the impetus provided by Professor Gould for this
paper, and his invaluable aid in pointing out the known results.
REFERENCES
1.
Paul S. Bruckman, "An Interesting Sequence of Numbers Derived from Various Generating Functions," |^bojiacJ_Quar^ely, Vol. 10, No. 2, 1972, pp. 169-181.
2.
RENEWAL NOTICE
Renewal notices, normally sent out to subscribers in November or December, a r e now
sent by bulk mail.
This means that if your address has changed the notice will not be for-
r
Let M denote the number of normalized (first row in natural order) r X n. Latin r e c tangles , M1 = 1 for all n. The M2 are the rencontre numbers [ l ] . Several methods are
available for computing the M3 [ l , 2, 3 ] . In this report, a procedure is presented that is
r
effective in finding M for r < 4.
CALCULATION OF M2 AND GENERAL FORMULA FOR M r
n
n
Consider the diagram (Fig. 1, drawn, as are all the diagrams, for n = 5)
of an n x n
consisting
Label each
cell with the numerical indices giving its position in the square, writing them as a one-column
matrix with the row index at the top, and mark those cells whose 2 indices a r e different.
Call the marked cells good, the others bad.
with all indices different.)
(For any r ,
Now expand the diagram into " t e r m s " by taking one cell from
each row and each column, keeping the row indices in natural order.
= n(k),
r, n
For r
>
Figs. 2 and 4), referred to in this report as the n r - c u b e , and obtain the formula, valid for all
r > 0.
n
M^ = ^ ( - l )
(1)
Ak
-
[(n-k)!]^1 .
k=0
denotes the number of k-tuples in the expansion of the n r - c u b e . (The word kr, n
tuple, in this report, will always mean an ordered set of k bad cells, no two of which are
Here A
(a,b,c) f
cells: a - c e l l s , having all 3 indices alike, and /3 -cells, having exactly 2 indices alike.
241
The
242
of
Kv
Wi
of
u\
x^
XXR " X
XX
X
X
XX
X X X XX X
XX x X X XXK
II
/
/" Oki
fi
"~"\
XX
X
XX
X X
XX
XX
XX
X
XX
X X X
X X
X
XX
XX
1 ^ ,
L/ 6
bar
stripe ^
v
X X^K
layer
Fig. 1 5 2 -Cube.
(Good cells outlined
and cross-hatched)
xmX X
<*>
[Oct.
fstripe
X
X
XX
XX
X X s XX
X
X
XX
X
X X
X X X
X
X X
X
XX
XX
X X X
X
X
X Xj
X X
X Y>
XX
X X
X
X
X
X
XX
v_
ybar
Fig. 4 5 -Cube (which is also the 54 - #-Cube.
4
1973]
243
Table 1
Formulas for the o>
for a\9
k = 0.
Except
all co
a r e 0 for k > 0 and for values of n > 0 not covered by the formulas.)
ak = n{ak-]
+ (n - l ) ^ ^ - * + 3 y k ^ + 6(n - 2 ) 6 k " ^ ] i
n
( n-1
L n-1
n-1
n-lj/
(n a 2)
(a{ -
(n a 1)
i-
^n = ^ n - k ( n - ^ [ 2 ^ 1
X k = 6 k - 2k
n
n
+ (n
"
(n ^ 3)
n-l]
(n ^ 1)
2)8
.k-1
Kl+ (n -2)
+
k-1
'n-1
k-1
n-1
(n s
2)
(n a
2)
k-1
n-1
+ (n - 3)^ = 1]}
(n - 3 ) [ * J
, k = ^ k - 3k(n - 2) [ 2 ^
(n 2= 2)
k+1
n+l
^ I i + ( n - 3 ) ^ : 1 + f g k-l>
:t>
(n ^ 1)
1)
(n > 2)
2 ^
(n - 3 w j
(n
3)
(n
2)
(n s
3)
(n a
4)
3(n-2)M k + , k ]}
244
e -cells.
[Oct.
(2)
'
(a*1 = n3 - n
Choosing ( 2 , 1 , 1 ) ' ,
we get the n 3 - j3 f -cube of Fig. 3. It differs from the n3 - a f -cube only in the first layer, all
the differing cells being /3 f -cells in the n 3 - a%-cubes.
Accordingly,
(3)
The factor k in the second t e r m on the right e n t e r s because of the way in which
k-tuple"
M is obtained.
n
CALCULATION OF M^
4
written in the order: stripe, b a r , row, column; the intersection of stripes a and b is called
the field [ a , b ] .
r = 3.
There
a r e now, besides the good o r e-cells, four kinds of bad cells: a - c e l l s , with all indices alike;
/3-cells, with exactly three indices alike; / - c e l l s , with two distinct p a i r s of like indices, and
S-cells, with exactly two indices alike.
The numbers of cells of each type a r e again easily found, and we have at once the formk
k
k
ula for A 4 > n = an given in Table 1. j3n is obtained from the n 4 -j3-cube resulting from
choosing ( 2 , l , l , l ) f in the (n + l) 4 -cube (second formula in Table 1), just as j3 ! n was obk
k
tained in the case r = 3. y and 5 are not so immediate, but can be found by the method
J
'n
n
outlined below.
We first analyze the n4 - /3-cube much as we did the n4 - a-cube. That is to say, we
study the types and topographical distributions of the second m e m b e r s of those p a i r s of cells
(n + l) 4 -cube whose first member is a selected j8-cell.
of the
4
(The cells in the first stripe all retain the same designations they had in the n4 - a-cube;
1973]
245
can
be obtained from the n - /3-cube (Fig. 5) by removing all the good cells from the field [ l , l ]
and inserting good cells at those places in the first bar indicated with o in the diagram. This
leads to the third formula in Table 1.
To get 6 n ,
the cube). This adjusted n - 5 - c u b e may be obtained from the n4 - j3-cube of Fig. 5 by r e moving all the good cells in
dicated with *.
[l,n]
and inserting good cells at the places in the n" 1 bar in-
246
Oct. 1973
i s found by considering all (rather than only the bad) cells of the n -cube,and p r o -
effective.
example.
All but ^ j ^ and 7r^ may be found in the following way, AjJ being used as
"new" good cells, marked x in Fig. 5 (i. e., cells that a r e good in the /3-cube but bad in the
a-cube).
The cell p a i r s that must be examined in this process all prove to be reducible to
in Table 1.
All the r e s t of the formulas of Table 1 except the last two a r e derived in the same way.
If the choice of the A-cell had been inappropriate ( e . g . , (2,2,3,3) f )>
would have met an impasse and haive failed.
cells,
the procedure
, ]|, , ^, 77j^.
The
latter expansion is analogous to that used to find a, above, n^ i s found by using all, r a t h e r
than only the bad, cells in the diagram, by analogy with the derivation of ^ .
There result
whose
values a r e : for n > 3 , all o> a r e 1; for n = 3, TT is 0 and all others are 1; for
n
o
n = 2, ejj, n\9 \ and ff| a r e 0 and all others a r e 1; for n *= 1, a, Z^0, y j and 8J a r e
1 and all the r e s t a r e 0, The o ^ can be checked by direct count in the appropriate diagram.
k
JL
F o r any given n > 0, all the A ,
can now be calculated, and IVT can be found by
substituting them i n ( l ) . Some enumerations of 4 X n Latin rectangles obtained by the method
h e r e presented a r e :
=
24
M| =
1,344
Mj =
393,120
Mj =
155,185,920
:M|
M$ = 143,432,634,240 .
REFERENCES
1. J. Riordan, An Introduction to Combinatorial Analysis, Wiley, New York, 1958, p. 57 ff
and p. 204 ff.
2.
P r o c . London
4.
G. Polya and G. Szego, Aufgaben und Lehrs'atze aus d e r Analysis, Dover, New York,.
1945, Problem VIII, 23, Vol. II, p. 120 and p. 327.
1.
INTRODUCTION
An ancient process for generating the sequence of prime numbers is known by the name
of The Sieve of Eratosthenes.
widely known.
In particular,
we can obtain the sequence of values for some of the common arithmetic functions. F i r s t we
shall discuss this Sieve of Eratosthenes and some of its modifications, then we will proceed
to some "sieves" for generating other sequences.
2. THE SIEVE OF ERATOSTHENES AND MODIFICATIONS
We recall that in o r d e r to obtain the sequence of p r i m e s by this method, the sequence
of integers g r e a t e r than 1 is first written down. Starting with 2 we then put a slash through
each second number beyond 2 in this sequence.
2 which is not crossed out, so that we then put a slash through every third number beyond 3.
(Note that, for example,
beyond 3 which is not yet crossed out, we next put a slash through each fifth number beyond
5, and continue in a similar manner. Those numbers remaining (not crossed out) are p r i m e s .
In o r d e r to place this p r o c e s s in a setting which is more suitable for generalization, we
will now modify the process in o r d e r to generate the sequence of values of what we shall call
the characteristic function for prime numbers, denoted by X p .
Xp(n) = 1 if n is a prime; X p (n) = 0 otherwise. In Table 1 the construction of the s u c c e s sive sequences is illustrated.
natural o r d e r which will thus indicate the position numbers for the elements of the sequences.
Table 1
Successive Sequences for X (n), N = 33
1
J2_
1
3 4 5 6 _7__8__9_ 0
A<>
0 1 1
A(1>
0 1
A(2)
A(3)
1 1
0 1 1 0
2
9
J-A JLJLJLJ>_JLJL A
1 1 1 1 1 1 1 i
_1_ J2_J3_
3
_8_
_9_
__0_
J7_
JLJLJL
I I 1 i I 1 1 1 1 1 I i
J_J2_ 3
I 1 1 1 1 1 1 1
1 0 1 0 1 0 1 0 i
0 I 0 I 0 1 0 1 0 1 0 I 0 1 0 1 0 1 0 1
0 0 0 i
1 0 0 0 1 0 i
0 1 0 0 0 1 0 I 0 0 0 1 0 1 0 0
0 1 1 0 1 0 1 0 0 0 1 0 I 0 0 0 I 0 1 0 0 0 1 0 0 0 0 0 1 0 1 0 0
248
[Oct.
Av
enter 0 in the first position and 1 otherwise for this first row of the table and for
The case N = 33 is illustrated.
we
nN,
position of the first non-zero element and denote this position by aj (= 2) and then convert
the entries in positions m a ^ m = 2, 3, 4,
m = 2, 3, 4,
(every a , - t h entry b e -
A<k>.
It is worth noting that, of course, the p r o c e s s can be terminated at A
^ N and the sequence A "
'
if
a, >
The reason that this termination is possible is that the smallest number which a, actually
sieves out is a,2, since a, a. for a. < a, has been sieved out at an e a r l i e r step.
F
k'
k j
j
k
In this construction the actual sieving out of the number n is indicated by the conversion of an entry 1 to an entry 0 in position n of the sequence. If a 0 has already appeared,
this indicates that the number had been sieved out at a previous step; that i s , the number
actually possessed a smaller prime factor than the number currently being used as the sieving number.
The entire process involves (1) the location of a non-zero element, (2) a count-
ing p r o c e s s , and (3) a change of entry. We note that no divisibility checks a r e used. One
further comment. This process does not involve using any of the sequences A^
for m <
(k-1)
(k)
k - 1 , but only the sequence A
to produce A . As a result, those preceding sequences
need not be saved. Even though the original p r o c e s s i s ancient, in this form it is quite adaptable for digital computers.
For some purposes a simpler sieve which yields slightly different information is of
value; this is illustrated in Table 2.
Table 2
F i r s t Modification,
N = 33
1
1 _2__3_
__J5_J3__7__8_JLJLj _ j _ J3___J5_J_
7 _8_
J^J)_JL
_2__3__4_
_6__7___8_J9_
J>_
J)_JL_J_J3_
A<>
1 1 1 1 1 1 1 1 i
I l
1 0 1 0 1 0 1 0 i
0 l
0 1 0 1 0 1 0 1 0 I 0 1 0 1 0 1 0 1 0 1 0 1
1 0 0 0 1 0 1 0 0 0 l
0 1 0 0 0 1 0 1 0 0 0 1 0 1 0 0 0 1 0 1 0 0
1 0 0 0 0 0 1 0 0 0 l
0 1 0 0 0 1 0 1 0 0 0 1 0 0 0 0 0 1 0 1 0 0
A(2)
A(3)
1 1 1 1 1 1 1 1 I 1 1 1 1 1 1 1 1 1 1 1 1
The change of the results i s indicated by the appearance of 1 in position 1 and in the
appearance of 0 in position p.
1973]
249
values of the characteristic function for the set which contains 1 and those p r i m e s satisfying the inequality N/N" < p k < N. The process is simpler in the sense that for the general
step we sieve out the numbers ma,
counting from the beginning.
a
>
'
if
^ '
These first two p r o c e s s e s have the disadvantage in that multiple sieving of elements
problem for the modified sieve, although it is a much more complicated procedure.
Consider the sequence A^
tion 1 which appears is in position a4 -(= 2) we sieve each second element by subtracting 1
from the entries in position ma, , m = l , 2, 3, 4, , then we rename the resulting s e (1)
(1)
quence A ' . Next, the first 1 which appears beyond position 1 in A
is in position a2
;
(= 3) and we use the sequence A
itself to generate the sequence of elements which a r e to
be sieved out of A^
final result of this second modification is the same as that of the first modification. The d e tails a r e illustrated successively in Table 3.
Table 3
Second Modification,
N = 33
3
2
1
__ 2 3
1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 12 3 4 5 6 7 8 9 0J
A^"
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
-1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1
1 0 1 0 1 0 10 10 1 0 10 10 1 0 10 10 1 0 1 0 1 0 1 0 1 0 1
-0
-0
-0
-1
-1
-0
-1
-1
-0
-1
-1
A< 2 >
1 0 0 0 1 0 10
-1
0 0 1 0 10 0 0 1 0 10 0 0 1 0 1 0 0 0 1 0 1 0 0
-0
-0
-1
-0
-0
1 0 0 0 0 0 10
0 0 1 0 10
A(3)
0 0 1 0 10 0 0 1 0 0 0 0 0 1 0 1 0 0
(k 1)
A useful way of thinking of the process is that A v
' is expanded by the factor a, and
subtracted from itself.
Another method closely related to this second modification and which eliminates m u l tiple sieving is discussed by G. S. Arzumanjan [ l ] . A geometric construction for the second
modification is given in [3].
250
[Oct.
Although these two modifications are not perhaps very important for the problem of
generating p r i m e s , they have been presented in detail because of their similarity to sieves
for other sequences which will be discussed in the next section.
3.
the case of primes we can think of the composites as falling through the sieve and being discarded; this present alteration can be thought of as collecting those things which fall through our
sieve in little boxes. In o r d e r to indicate how this process goes, let B ( 0 ^ be the sequence with
0 in each position.
To produce B^
. The first 0
which appears beyond position 1 of B ( 1 ' is in position p 2 (= 3). We next add 1 to each entry in B
in position mp 2 , m = l , 2 , 3 , 4 , to produce B
ner we can state the general step of this iterative process. Locate the first 0 which appears
, this will be in position p k - Next add 1 to each entry in B ^ " 1 ^
beyond position 1 in B*
in position mp, , m = 1, 2, 3, 4, for mp, < N. This will produce the sequence B ( k \
The process can be stopped at sequence B l "
the entry in position p fe , will be altered. The remaining entries which are 0 and a r e beyond
position 1 now indicate p r i m e s satisfying N/2 < p < N , if N > 4, and hence the process can
be completed by converting the 0 entries to 1 in these positions N/2 < n < N of B ^ " 1 ^ . The
(k)
sequence B v
coincides with the sequence for u for n < N. The results for N = 33 are
given in Table 4.
Table 4
Sequences for u(n), N = 33
2
^or
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
B (
0 1 0 1 0 1 0 1 0 i
B (3)
1 1 0 3 0
11
0 1 1 1 1 2 0 1 i
0 1 1 1 1 2 1 1 i
0 1 1 1 1 2 1 1 i
0 1 1 1 1 2 1 1 i
2 0 2 0 1 2 1 0 2 0 2 1 1 0 2 1 l
11
2 0 2 0 2 2 1 0 2 0 2 2 1 0 2 1 l
1 2 0 3 0
11
1
1 2 0 3 0
12
2 1 2 1 2 2 1 0 2 0 2 2 2 0 2 1 2 1 2 0 3 0
12
2 1 2 0 2 2 1 0 2 0 2 2 2 0 2 1 l
1
B (6)
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 l
1
B (5)
10
1
1
B (4)
0 1 0 1 0
1
1
0 1 1 1 0 2 0 1 i i 0 2 0 1 1 1 0 2 0 1 1 1 0 2 0 l 1 1 0 2 0
1
1
1
1
1
i
1
B(2)
n Qf-nr
P onjp
1 1 2
1973
251
at the k
factor p ^ of n exactly once. Hence we have generated the sequence of values for the function y; that i s , u(n) appears in position n,
It now becomes an analogous exercise to obtain a sieve p r o c e s s for computing the s e quence of values r(n),
If we next attempt to compute the values Q(n), the total number of prime factors of n,
the procedure seems to become more complicated.
lows, starting with the sequence CP ' for which all entries a r e
0.
In o r d e r to construct
22,
etc. , until
> N.
counting the factors 2 reminds one of the second modification in Sec. 1, if we consider that
the first subsequence c\
to be added to C
successively c }
n,
which is to be added
,#e
to C
2m,
, etc.
divides
Table 5
Sequences 3 f o r Q ( n ) , N == 33
2
1
1 _2__3_
^1
1 2
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
CW
0 1 0 2 0 1 0 3 0 1 0 2 0 1 0 4 0 1 0 2 0 1 0 3 0 1 0 2 0 1 0 5 0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
c< 2)
0 1 1 2 0 2 0 3 2 1 0 3 0 1 1 4 0 3 0 2 1 1 0 4 0 1 3 2 0 2 0 5 1
1
1
1
1
1
1
1
c(3>
0 1 1 2 1 2 0 3 2 2 0 3 0 1 2 4 0 3 0 3 1 1 0 4 2 1 3 2 0 3 0 5 1
1
1
1
1
c(4)
0 1 1 2 1 2 1 3 2 2 0 3 0 2 2 4 0 3 0 3 2 1 0 4 2 1 3 3 0 3 0 5 1
1
1
1
c(5)
c(6)
0 1 1 2 1 2 1 3 2 2 1 3 0 2 2 4 0 3 0 3 2 2 0 4 2 1 3 3 0 3 0 5 2
1
1
0 1 1 2 1 2 1 3 2 2 1 3 1 2 2 4 0 3 0 3 2 2 0 4 2 2 3 3 0 3 0 5 2
I
a
i 0
Qf
"** & t o p
1 1 2 1 2 1 3 2 2 1 3 1 2 2 4 1 3 1 3 2 2 1 4 2 2 3 3 1 3 1 5
252
[Oct.
from C
Then C
to C . C2
could also be constructed from c }
(2)
from position m of C\
and 0 otherwise.
by entering in position
C2
mp 2 the entry
Consider
(k 1)
and
locate the first 0 entry beyond position 1; this will be in position p. by analogy to the
(k)
sieve for p r i m e s . To begin the subprocess, we form Cj
by entering 1 in position mp, ,
(k)
>
m = 1, 2, 3 , 4, , until mp.
N. The subsequence C:
i s formed from the subse(k)
1
quence C:
by entering in position mp, , m = l , 2, 3, , the entry from position m
of Ci i and 0 otherwise until mp? > N. The sequences Care successively added to
c ( k - l ) to produce C ^ .
ed separately and then added, the addition p r o c e s s , of course, can be c a r r i e d out as one
p r o g r e s s e s . ) The process can be stopped at the same point in the computation as for the
values of o and the remaining 0 entries converted to 1, using the same reasoning.
It is
not difficult to see that we actually have obtained the values Q(n).
A slight modification of the entries leads to the sequence
function of interest in the theory of numbers.
X(n) = (-1)
another
comes an exercise.
4. THE NUMBER OF REPRESENTATIONS OF A NUMBER
AS A PRODUCT OF NUMBERS CONTAINED IN A GIVEN SET
We shall next consider the following problem. Given a subsequence S of natural numbers 1
<
&i < a2 < ' * , either finite or infinite, we wish to compute the number of distinct
has been generated separately and we let R(n) denote the number of such representations of
n.
the sequence Q.
We let R
position dii of R
tion 2a2 of R
entry at position k of R
>
1 of R
is
1973]
entered in position 1 of R
from position la 2 of R
in o r d e r for m = 1, 2, 3, ,
position ma 2 of R
R
253
R '
from
R^
we
-j \
' if
a, > N.
In Table 6 we have
chosen S = \ 2 , 3 9 4 9 5, 129 309 72/ and we indicate the steps of the computation.
Note
the iterative process which occurs within the computation for each sequence.
Table 6
R(n) for S = { 2 , 3, 4 9 59 12, 30, 72} 9 N = 33
I
2
3
1 JLJJ_J_JLJLJLJ3_JLJLJLJLJLJLJLJLJLJLJLJLJLJLJLJLJLJLJLJ3_JLJLJL_JLJL
H (0)
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
R(1)
1 I 0 1 0 0 0 1 0 0 0 0 0 0 0 I 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
R(2>
1 l 1 1 0 I 0 1 i 0 0 i 0 0 0 i 0 1 0 0 0 0 0 1 0 0 1 0 0 0 0 1 0
R (3)
R (4)
1 I 1 2 0 l 0 2 I 0 0 2 0 0 0 3 0 1 0 0 0 0 0 2 0 0 1 0 0 0 0 3 0
1 I 1 2 1 i 0 2 I i 0 2 0 0 1 3 0 1 0 2 0 0 0 2 1 0 1 0 0 1 0 3 0
H (5)
1 I 1 2 1 I 0 2 i l 0 3 0 0 1 3 0 1 0 2 0 0 0 3 1 0 1 0 0 1 0 3 0
1 i 1 2 1 i 0 2 i l 0 3 0 0 1 3 0 1 0 2 0 0 0 3 1 0 1 0 0 2 0 3 0
The iterative procedure from R
to R
(k)
(n) = R
(k_1)
(n) + R
if n m a k ,
(k)
(m) 9
if n = ma fe .
a. = p, ,
then, of course,
R(n) = 1
since
254
Oct. 1973
a r e given in a recent paper by C. T. Long [7]. Beginning with V. Brun [2.] techniques involving double sieving and other modifications have been used to study the twin prime probl e m , the Goldbach conjecture, and other problems. An interesting article by David Hawkins
[ 5] on the sieve of Eratosthenes, random sieves, and other m a t t e r s is well worth consulting.
REFERENCES
1. G. S. Arzumanjan,
ff
R. G. Buschman, "Pencils of Rays and the Sieve of E r a t o s t h e n e s , " submitted for pub.
4. V. Gardiner, R. L a z a r u s , M. Metropolis, and S. Ulam, "On Certain Sequences of Integers Defined by Sieves, " J l a t i ^ M a g . 29 (1955/56), pp. 117-122.
5. David Hawkins, "Mathematical S i e v e s , " Sci. Amer. 199 (Dec. 1958), pp. 105-112.
6. D. Hawkins and W. E. Briggs, "The Lucky Number T h e o r e m , " Math. Mag. 31 (1957/58),
pp. 277-280.
7. C. T. Long, "On the Moessner Theorem on Integral P o w e r s , " Amer. Math. Monthly, 73
(1966), pp. 846-851.
8. A. Moessner,
INTRODUCTION
This paper deals with the discussion on the graphical aspects of the Fibonacci numbers
through the topological index [l] which has been defined by the present author for nondirected g r a p h s . 1
A graph G consists of points (vertices or atoms) and lines (edges or bounds) [ 2 , 3 ] .
We are concerned with such connected non-directed graphs that have no loop (a line joining
to itself) and no multiple lines (double or triple bonds). An adjacency matrix A
for graph
) 1
I 0
The m a t r i x character is independent of the way of the numbering of the points. A c h a r a c t e r istic polynomial or a secular polynomial P(X) is defined as 2
(2)
i=0
where E is a unit matrix of the o r d e r N and X is a s c a l a r variable.
Consider a s e r i e s of path progressions { S N } ,
for which
P(X)
can be expressed as
(see [4])
(3)
P(X)
^E^^k^"
" '
following page.
O r i g i n a l l y this idea came out quite independently from other works especially published in
mathematical journals. However, thanks to the communications from the colleagues in this
field, several important papers were found to be relevant to this problem. In this paper the
relevant papers will be cited as many as possible.
A l t e r n a t i v e definition
(2f)
PCX) = (-1)
det | A - XE |
256
[Oct.
Table 1
/
N
G(S N )
, \
(" ;
k == 0
]C p ( G ' k )
1 = p(G,k)
k=0
"
1
= f
1
2
3
5
.1
13
7
8
10
21
15
10
34
On the other hand, from the combinatorial theory we know the following relation under
the name of Lucas (see 5, 6 ] ) .
m
(4)
N
IN
2-r \
) '
k=0
where fN
is the N
f0 = h = 1, *
(5)
f
N-l
+ f
(N
N-2
'3'-">
'
The sums of the absolute values of the coefficients of the characteristic polynomial for the
graph \ S N } form the Fibonacci s e r i e s .
by 45 degrees, and we get the P a s c a l ' s triangle or the pyramid of binomial coefficients,
from which the Fibonacci numbers can be obtained by adding the coefficients diagonally (just
the r e v e r s e of the above p r o c e d u r e ! ) .
(See [ 7 ] . )
) - Consider a group
Choose an a r b i t r a r y set of k
points (black c i r c l e s ) , place k additional points (crosses) one-by-one below them, and join
all the N points together by drawing consecutive N - 1 lines as in Fig. l b to get a path p r o gression with N points, or N - 1 lines. This means that the value I
I is the number
of ways in which k disconnected lines (vertical lines in Fig. lb) are chosen from graph L..
* Alternative definition can be used as fi = f2
1-
1973]
257
N-k points
(a)
N points
VV
(b)
(v)
Define
a non-adjacent number p(G s k) for graph G as the number of ways in which k disconnected
lines a r e chosen from G. A Z-counting polynomial Q(Y) and a topological index Z a r e d e fined respectively as
m
(6)
p(G,k)Y k
Q(Y) = J^
k=0
and
m
Z = Xp(G'k)
(7)
= Q(1)
'
(See
WB)
k=0
Note that for the s e r i e s of path progressions { S N } in Table 1, the p(G,k) number is nothing
else but I
"
t e r m in the
o r for brevity
(80
= fN
S N - fN
F u r t h e r , for any tree graph with or without branches but with no cycles, the following
relation can be proved by t e r m - t o - t e r m inspection of the expansion of det ,|A+ XE | intoP(X):
(9)
P(X) = J ] ( - l ) k p ( G , k ) X N " 2 k
(G G Tree) .
k=0
and
numbers
have been published for smaller tree [8] and non-tree [9] graphs. F o r non-tree graphs, Eq.
(9) no longer holds but P(X) can be expressed as the sums of the contributions like the right-
258
[Oct.
Table 2
G
k = 0
]p(G,k)
-J_
!
1
_U_
10
_ ! _ _ _
11
I _ _
_ ' _ _
12
nomial [12, 13] the topological index does not uniquely determine the topology of a graph.
However, it is generally observed that the Z value gets s m a l l e r with branching and l a r g e r
with cyclization.
F o r evaluating the
roughly
Z values of l a r g e r and
They can be proved by
(See [ l , 10].)
COMPOSITION PRINCIPLES (CP)
L and M.
Then the
G = LXM + A X B X C X D X E X F .
F o r applying this principle there i s no restriction in the number of subgraphs incident to the
chosen line i ,
since the Z values of a point graph (Sx) and a vacant graph (S0)
unity,
(11)
So = ^
= 1.
N-1
N-2 '
a r e both
1973]
<S>
(ffi
( <
SM-
ran
260
[Oct.
Comparison of E q s . (11) and (12) with Eq. (5) yields Eq. (8).
For graph S N with even N (= 2n) we get the relation
S 0 = S2 + S2 2n
n
n-1
(13)
(14)
Similarly, we get
(15)
S 0 x l = S (S ^ + S
J
2n+l
n n+1
n-1
or
(16)
f ^ = f (f J . 1 + f n ) .
2n+l
n n+1
n-17
Corollary to C P 1 . If the line to be deleted is a member of a cycle, the deletion gives
G = L + M .
By use of this corollary the Z values for the s e r i e s of N-membered cycles (N-gon,
N =
N-2
and the s e r i e s of these Z values form what a r e known as the Lucas sequences { g N } ; namely,
(see [5])
C
(19)
gi = 1.
S2 = 3
+ g
N-l
N-2 '
% = fN
hl-i
From the correspondence relation of Eq. (19), a monogon and a digon may be defined,
spectively, as a point graph Ct
(= St)
re-
1973]
261
Table 3
N
G(C N )
2
3
4
5
6
7
8
o
A
O
O
O
O
p(G 5 k)
1 2
k =0
1
1
11
18
14
29
20
16
47
from it a point p. The number of the lines incident to poiiit p should be at least two but not
n e c e s s a r i l y be six as in this example,,
In this c a s e , we
d9 e and f in
G, and we get subgraphs D, E , F and L. (4) Delete both of the groups of lines a, b , ,
f in G, and we get subgraphs A, B , , F . With these subgraphs we have
(21)
G = A X B X C X M + D X E X F X L - A X B X C X D X E XF .
Composition Principle 3 (CP3). Further consider a graph G in Fig. 4a in which two
(1) Delete a line from S3 and rejoin the two resultant subgraphs to get L.
lines.
m o r e line from S2 in L and rejoin the subgraphs to get M. The Z value for G i s given
by (see Fig. 4)
(22)
G = L +M.
This i s also applied to" the case in which A and B a r e joined with two paths to form a
cycle to give the relation (19) (Fig. 4b).
RECURSIVE SEQUENCES
A recursive sequence { a ^ } i s defined as
262
a)
<J
lo
CD
r-4
d
rH
PK
o
O
S
o
u
S
(CD)
> Co:
> Co
1973]
(23)
263
k
C
*N E i V i
i=l
Both of the Fibonacci and Lucas sequences a r e the special cases with C 4 = C 2 = 1, C. = 0
(i
>
2) but with different initial conditions. One can find a number of graphical s e r i e s whose
topological indices form recursive sequences as in Fig. 5. They can be proved by the composition principles.
topological index.
The most important point in this discussion is that a number of relations in the r e c u r sive sequences can be inspected and proved by applying the composition principles to the
graphical equivalent of the sequences.
APPLICATION TO CHEMISTRY
Let us confine ourselves to a class of chemical compounds, saturated hydrocarbons,
whose topological structure is expressed as a structural formula.
An example is shown in
Fig. 6 for 2-methylbutane (a). Since carbon (C) and hydrogen (H) atoms, respectively* have
t e t r a - and mono-valencies, for describing the whole structure only the carbon atom skeleton
(b) is sufficient, which is equivalent to graph (c). The s e r i e s of graphs in Table 1 a r e read
in chemical language as methane, ethane, propane, butane, etc.
normal paraffins.
Thus the topological indices of normal paraffins are shown to form the
Fibonacci sequences.
entry in Table 2) both have Z = 5. However, it was shown that the topologi-
cal index can be used as a rough sorting device for coding the complicated structures of
chemical compounds [14].
It was also shown that the topological index of a saturated hydrocarbon is correlated well
with some of the thermodynamic quantities such as boiling point through its entropy, which i s
a m e a s u r e of the degree of freedom in internal rotations of a flexible molecule [15].
Characteristic polynomials appear in the application of quantum mechanics to the study
of the electronic structure of molecules. The simplest method is the H'uckel molecular orbital
method, in which the problem is reduced to obtaining the solution of a secular equation
P(X) = 0 (see [10, 16, 17]).
SYNOPSIS
Define a topological index Z as the sum of the non-adjacent number, p(G,k), which is
the number of ways in which such k disconnected lines are chosen from graph
G.
The
values for the path progressions { S N } form the Fibonacci sequences, while those for the
264
[Oct.
CO
CJ>
GO
5H
CO
in
CM
CO
O
CD
&
02 II
-a
in
CSI
CD
Is
CO
I
II
oP
CO
CM
CM
in
CM
1973]
265
HC-H
HC
CI
I
H
H
I
CH
H
(b)
(a)
(c)
Fig. 6 Structure and Graph of 2-methylbutane
s e r i e s of cycles { C N } the. Lucas sequences.
4.
5. H. S. M. Coxeter, Introduction to Geometry, Chap. 11, John Wiley and Sons, Inc. ,
New York (1965).
6. J. Riordan, Combinatorial Identities, p. 89, John Wiley and Sons, Inc. , New York
(1968).
7. N. N. Vorob*ev, Fibonacci Numbers, Nauka P u b l . , Moscow (1964).
8. K. Mizutani, K. Kawasaki and H. Hosoya, "Tables of Non-Adjacent Numbers, C h a r a c teristic Polynomials and Topological Indices.
ig7
12. A. T. Balaban and F . Harary, "The Characteristic Polynomial Does not Uniquely Determine the Topology of a Molecule," J. Chem. Doc. , 11, pp. 258-259 (1971).
13.
14. H. Hosoya, "Topological Index as a Sorting Device for Coding Chemical Structures,! 1
J. Chem. D o c . , in p r e s s .
15. H. Hosoya, K. Kawasaki and E. Mizutani, "Topological Index and Thermodynamic P r o p e r t i e s , I. Empirical Rules on the Boiling Point of Saturated Hydrocarbons," Bull. Chem*
Soc. Japan, submitted.
16. E. Huckel, Z. physik, 70, p. 204 (1931); 76, p. 628 (1932).
17. A. Streitwieser, J r . , Molecular Orbital Theory for Organic Chemists, John Wiley and
Sons, Inc. , New York (1961).
1971. 138
$5.95.
The I Ching Games, whose names translate as "The Wisdom Plan" and "The Beneficial
to Wisdom P l a n , " a r e considered among the most important ever written, since they a r e
thought to improve the player's ability to learn while advancing his psychological development.
The first game is also called the Tangram, being a seven-piece dissection of a square into a
s m a l l e r square, five isosceles right triangles, and a parallelogram,
assembled into an infinite variety of recognizable pictures.
which can be r e -
a square to also include the c i r c l e , and the problems become jigsaw puzzles of a thousand
delights.
The authors, as well as hoping to re-introduce the Tangram game and introduce the 15game to the West for the first time, give a history of the games and describe their relationship to the I Ching, the ancient Chinese Book of Change, the oldest book now known.
ABSTRACT
Let (a ls , a k ) = a denote a vector of numbers, and let C(u,n) denote the n X n
cyclic matrix having (a ls , a k , 0, , 0) as its first row. It is shown that the sequences
(det C(a,n) : n = k9 k + 1, . . )
and
(per C(a,n) : n = k, k + 1, )
(-1)
s i s n 7T
where 77 is a
permutation.
INTRODUCTION
While she was a student at Lowell High School, Beverly Ross [2] generalized an e x e r cise given by Marshall Hall, J r .
or
SDR f s
for short.
8 e
, am
Let
with a. = A.
e r c i s e is the case m = 7 of the following problem posed and solved by Ross: Let A. = { i ,
i + 1, i + 2/ denote a 3-set of consecutive residue c l a s s e s modulo m
!
for
i = 1, , m.
n_i
+ L n _ 2 for
For example, it follows from this result that the solution to Hall's exercise
is 2 + L 7 = 31.
In this note we give a new proof of Ross 1 theorem, and indicate a generalization.
ROSS1 THEOREM
We shall require a simple result which appears in Ryser [3]; namely, the number of
!
SDR s of an m-tuple B* = (B l s , B m )
the incidence matrix of 13. Since this fact is an immediate consequence of definitions, we
give them h e r e .
[b(i, j)]
Bl9 , B m
of "B = (Bt, , B m )
is
defined by
i 1 ,
b ( i
for i = 1, , m and j = 1, * , n.
J0,
if
j G B. ,
if
j$B.
fined to be
267
[r(i,j)]
is d e -
268
[Oct.
p e r [r(i,])]
= V \ ( i , 7 r l ) r ( 2 , t f 2 ) r(m,7Tm) ,
where the index of summation extends over all one-to-one mappings 77 sending \ 1 , , m}
into ( l , * , n } .
The incidence matrix C
considered by Ross is an m X m
of sets Aj, , A
( 1 , 1 , 1 , 0 , * ,0);
that i s , the first row has its first three components equal to 1 and the r e s t of its components
equal to 0.
cf =
= 2+L
. To do this, we define
three sequences of m a t r i c e s :
1
1
1 1 1 '
D3
D4
1 0
0
1
1
1
1
Fa =
1 1
0
0
1 1
0
F, =
0
0
D,
1 1
1 1
0
1 0
Eo =
"
FR =
[1
1 1
[1
1973]
269
the sum of the permanents of the minors of the l f s in a row or in a column of the matrix.
Second, the permanent of a matrix is unchanged by permuting the rows or by permuting the
columns.of the matrix.
equal to 0. Fourth, the permanent of a square matrix is equal to the permanent of the t r a n s pose of the matrix.
of C
Expanding per C
, we find
c
(1)
m =
2d
m-l
+ e
m-l
= 4,5, )
Expanding per D
in t e r m s of the minors of the l f s in the first column of D , we find
m
m
(2)
., + f,. _,
m-1
m-1
(m = 4, 5, )
'
(3)
(4)
(m = 4,J 5,' )
' '*
m-1
2+F
+ F
m-1
- +F ^
m-1
m+1
2 +L
for m
^ , where F
m+1
m
= 1 + F , and c
3, 4,
A GENERALIZATION
Let a = (a ls , a^) denote a k-tuple of numbers and let T denote a k X (k - 1)
matrix having all of it's entries in the set {o, a.u , a ^ } . F o r each n ^ k define an n X n
matrix C(T,n) as follows:
n1
C(T,n) =
a4
a
T2
ai
Tj of T in the upper
right corner, and the lower triangular half T 2 of T in the lower left corner.
t r i e s in the first k - 1
columns of C(T,n)
are
0.
270
Oct. 1973
as its top row, we expand per C(T,n) by the minors of elements in the
top row of C(T,n). It turns out that these minors always have the form C(T., n - 1) where
T. is a k x (k - 1) matrix having all its entries in
(o, a 4 , , a ^ } .
such that
k
(1)
p e r C ( T , n ) = " / t . p e r C(T., n - 1) ,
i=l
where t
= a, .
(-1)
We have an equation like (1) for each matrix T; hence, we have a finite system of equations if we let T range over all possible k X (k - 1) m a t r i c e s with their entries in {o, a l9
, a ^ } . The existence of this system of difference equations implies the existence of a difference equation satisfied by the sequence
matrix T.
(per C(T,n) : n = k, k + 1, )
(det C(T,n) : n = k, k + 1, ) . )
A conse-
quence of the foregoing is the result proved by Ross, but evidently much m o r e is true.
Let rl9 , r n denote natural numbers with
natural number m ^ k define the collection 3"
1 = rA
<
< r n = k,
= {A1S , Am}
of sets A i
of residue
c l a s s e s modulo m where
A. = { r t + i, , r n + i } .
Let a(m) denote the number of SDR ! s of A
The proof of
this fact follows the proof of Ross Theorem given in the preceding section.
Note that our existence theorem has a constructive proof, but we do not have an explicit
expression for a difference equation satisfied by the sequence (per C(T,n) : n = k, k + 1, ) .
This gives r i s e to a host of interesting r e s e a r c h problems.
equation satisfied by the sequence (per C(k,n) : n = k, k + 1, ) where C(k,n) is the cyclic n x n matrix having as its first row (1, , 1, 0,. " , 0) consisting of k
l T s fol-
lowed by n - k 0 f s.
REFERENCES
1. Marshall Hall, J r . , Combinatorial Theory, Blaisdell Publishing Company,
Mass. , 1967.
2.
Waltham,
Beverly Ross, "A Lucas Number Counting Problem," Fibonacci Quarterly, Vol. 10 (1972),
pages 325-328.
But, for c e r -
tain c l a s s e s of polynomials, the roots can be derived by using hyperbolic trigonometric functions.
Here, we solve for the roots of Fibonacci and Lucas polynomials of degree n.
The Fibonacci polynomials F (x), defined by
F t (x) = 1,
F2(x) = x,
Li(x) = x,
L n + 1 (x) = xL n (x) + L ^ f c s ) ,
/nX
(1)
x + 's/ x 2 + 4
s
,
x Q
P
= ~
N/ x
+ 4
(2)
_n
x
n(
> = n + ^
n
F"n 1(x)
= 2-^4
"'
a - B '
The first few Fibonacci and Lucas polynomials are given in Table 1.
x = 1,
F (x) = F
sinh z = (e
- e" ) / 2 ,
cosh z = (e
271
+ e" )/2
272
[Oct.
Table 1
Fibonacci and Lucas Polynomials
F
n
1
2
L (x)
n
n
1
x* + 2
X3
+ 1
+ 2x
x 3 + 3x
x 4 + 4x 2 + 2
X4
+ 3x2 + 1
x 5 + 5x3 + 5x
X5
+ 4x 3 + 3x
X6
X7
x 6 + 6x4 + 9x2 + 2
2
x 7 + 7x5 + 14x3 + 7x
+ 5x + 6x + 1
+ 6x5 + 10x3 + 4x
6
+ 7x + 15x + 10x + 1
X8
sinh iy = i sin y
F (x) =
n
a
Then,
-g
fln
- P
zn
/ v
n , 0n
L (x) = a + j8
T
= e
lX n
-z <-
-nz
+ e"z
nz , / lV n -nz
+ (-1) e
Thus
p
/v\
2nw
sinh 2nz
coshz '
, ,
2n+lw
cosh (2n + l ) z
cosh z
(3)
L 2 / v = 2 cosh 2nz,
Now, clearly the polynomial equation equals zero when the corresponding hyperbolic
function vanishes. For z = x + iy (see [ 2 ] , p. 55)
I sinh z | 2 = sinh 2 x + sin 2 y
|cosh z | 2 = sinh 2 x + cos 2 y
Thus, since for real x,
.
this implies that the zeroes of
sinh z a r e those of sinh iy = i sin y, and the zeroes of cosh z a r e the zeroes of cosh iy =
cosy.
be zero.
F (x) and
L (x)
to
1973]
cos y i
273
Thus,
x = + 2i sin k77/2n,
so that
sin 2ny = 0,
k = 0, 1, 2, , n - 1.
Specifically, the zeroes of F6(x) are given by x = +2i sin k77/6, k = 0, 1, 2, so that x =
F6(x) = x(x2 + i)( x 2 + 3),
working.
F2
1 (x)
= 0
only if
Then,
cosh z f 0,
i(2k + 1)77
(2n + 1)2 '
ly
so that
,0.
/2k
x = +2i sm I
+ l\
77
J rr ,
-,
k = 0, 1, , n - 1 .
F 2 n (x) = 0 :
x = 2i sin ,
F2n+1(x) = 0 :
x = 2i sin f | ^ 4
L 2 n (x) = 0 :
x = 2i sin \I 52n
7 j) ' -2 :,
2n+l(x)
0 :
k = 0, 1, , n - 1
/2k + l \
2 i s i n
77
J ^ :,
"^2n + i y
/2k + l \
k = 0, 1, , n - 1
k = 0, 1, , n - 1
k7T
2^TT'
k = 0, l , . . - . n - l
N/ x + 4 = 2i sinh z,
and a = i e , /3 = i e " .
^ . v
(4)
so that
Then, by (2),
- / zn
-zn\
/
1) sinh nz
.n-11 e
- e
\ _ .(nsinh z
\ e - e
/
L (x) = enz
n
+ e"nz = 2-incoshnz
^ 0. Since sinh nz = 0 if and only if sin ny = 0 or when z = iy, we must have ny = k77
so that z = ik77/n. Since i cosh iy = i cos y, x = 2i cosh z = 2i cos k77/n, k = 1, 2,
n-1.
Now, for the Lucas polynomials,
when ny is an odd multiple of 77/2,
z = iy,
cos ny = 0,
so that x = 2i cosh z
or
becomes
274
x = 2i cos
n
L (x) = 0 :
n
x = 2i cos
Oct. 1973
k = 1, 2, * * , n - 1
(2k
1)7T
2n
k = 0, 1, , n - 1
aXi
F (x) = 0, then the roots of L (x) a r e those roots of F 9 (x) which a r e not roots of F (x)
m
uW
n
n
Please note how this agrees with our r e s u l t s :
F 2 n (x) = 0
x = 2i cos ^
k = 1, 2, ' ' , 2n - 1
F n (x) = 0
x = 2i cos %& ,
j = 1, 2, , n - 1
L n (x) = 0
x = 2i cos
(2j + 1)7r
2 n
j = 0, 1, , n - 1 .
REFERENCES
1. Marjorie Bicknell, TA P r i m e r for the Fibonacci Numbers: P a r t VII An Introduction to
Fibonacci Polynomials and Their Divisibility P r o p e r t i e s , " Fibonacci Quarterly, Vol. 8,
No. 4, October, 1970, pp. 407-420. Also available as p a r t of A P r i m e r for the Fibonacci
Numbers, The Fibonacci Association, Dec. , 1972.
2.
Ruel V. Churchill, Complex Variables and Applications, McGraw-Hill, New York, 1960.
3.
RENEWAL NOTICE
Renewal notices, normally sent out to subscribers in November or December, a r e now
sent by bulk mail.
This means that if your address has changed the notice will not be for-
1.
INTRODUCTION
Generating functions provide a starting point for an apprentice Fibonacci enthusiast who
would like to do some r e s e a r c h . In the Fibonacci P r i m e r : P a r t VI, Hoggatt and Lind [l] d i s cuss ordinary generating functions for identities relating Fibonacci and Lucas numbers. A l so, Gould [2] has worked with generalized generating functions.
M\
j.
\A
v3
V^
n=0
Then
e
l + j? + - g j - + ~ ^ j - +
4J
(2)
^-e^ = ( l - l )
^^
x1
= F
n+1
+F
n
nf
i24^
Fn
is the n
^ - ^
...
n-1
known that
F n = (<*n - pn)/(a
(3)
- j3) .
fit
a - P
F^
F2t2
F81?
1!
2!
3!
F 4 t^
1!
4!
*
~J
n=l
275
n
n n!
then it is well
276
F 0 -^
[Oct.
(4)
at
i3t
,n
-j
=y
a - /3
F L
Z-*
n=0
n n!
(5)
If the power s e r i e s for e
and e
is
oo
(6)
+ e
= > L -7
Zt n n!
n=0
F o r a novel use for these elementary generating functions, the reader is directed to [3]
for a proof that the determinant of eQ
3.
is e
where Q = ( -
J .
L^J
tn
i,.
n n!
^
tn
B(t) = > b ^ ,
'
Zi n n!
n=0
and
n=0
it follows that
= E ( EuKVklnT '
A(t) B(t)
n=0 \ k=0
'
(7)
A(t)B(-t)=E(L(-l)n-k(nkKUS
n=0 \ k=0
= 1 for all n,
/
A(t) e
^ '
and
n
'-E s(jk)S
f
n=0 \ k=0
'
1973]
277
A(t
>
oo
T-^ n t n
tn
\-^
E If * ^ = E If
n=0
n=0
Then
n
/
A(t)B(t) =
~
n=0 \ k = 0 X
'
-(j>&*((!)-(i>)&*((!)-(0-(;>)i+oo
= 0+
4t2
,. 2t
+ e +te
v^
oo
t(2t) n
E -V-
V*
n=0
2ntn+1
-sr
n=0
V
(n + D 2 n t n + 1
- La
(n + 1)!
n=0
OO
(na^V
n!
(")
(A = n!
k!(n - k)f
1kI
4.
Generating function (4) and algebraic properties of a and /3, the roots of x 2 - x - 1 =
0, give us an easy way to generate Fibonacci identities. Useful algebraic properties of a =
(1 + is/T>)/2 and 0 = (1 - \l"5)/2
a2 = a + 1
. o?/3 = - 1
a - j3 = ^ 5
a
Take
j_ o
-,
a + p =
B(t) = e t
include:
a F
HI
m-1
- 0).
+ ^n
r
Then their
278
e (a+i)t
_ e(P+l)t
Qah
a - p
[Oct.
_ J$h
or - 0
(8)
zLr
n=0
On the left, we used s e r i e s property (7).
algebraic properties of a and p9
tJn
22n
n n!
k=0
'
n=0 \ k=0
then
(9)
-/3t
-at
2 ^ ("1}
S^T/5
,n
n nT
'
n=0
The identity resulting from (9) is
2 <Wi)
k=0
F. = ( - l ) n + 1 F
k
n
'
B(t)
and A(t)
as simple functions in t e r m s of
powers of e. Follow algebra as outlined in Eqs. (1) through (7), and equate coefficients of
n
-t
cP't j3^t
t / n ! The reader is invited to use B(t) = e" and A(t) = (e
- eH )/(a - p) to derive
FO1
2k
= F
k=0
For an identity relating Fibonacci and Lucas numbers, let
A(t) = (eat - ePt)/(a
- p) ,
B(t) = e* + ePt
1973]
279
Since B(t) is the generating function for Lucas number coefficients (see Eq. (6)), computing
the s e r i e s product A(t) B(t) gives
00
<">
/
/
n
/
\
\,n
r L
oo
2at
281
ji
2 ( 2 ( 0 * - t h - '-^r-- E **. r
n=0 \ k=0
n=0
yielding
. = 2nF .
k
n
k=0
e2at
n=0 \ k=0
(11)
+ e2^t
2et}
'
= ^ l ( 2
- 2 ) L
n=0
n
E(k)FkFn-k=l(2\-2> '
k=0
at
Bt
+ eK to derive
k=0
ni,
fc
- e^
nin,
and e
*)/( - /3),
, such as
B(t) = e*
+ eP
_ m,
\t n _ e2a
l X^fA
F
I Z-rl k J mk mn-mk)nT
n=0 \ k=0
/
- eP
00m,
- e2/3
STTjg
_V
^
n=0
tn
n
2
m n nT
280
V [y/n\ F
Z - / 1 Z-* I k /
n=0 \ k=0 X '
[Oct.
nHiA 2
nr
V - h *_f '
mk mn-mk I n !
/
1
a - p
\
- n m,
/ m 1/a iiiv,
00m,
= i C e 2 0 t + e2^ t - 2 e ( a ^
>*)
(nl)
OO
= y;i(2nL
- 2 L n ) ^
A/ 5
mn
m n!
n=0
m,
fim
The generalization of (12) found by A(t) = B(t) = e^
+ e^
is
00
/
/
n
/
\
\
,n
m,
0m,
2
*
n=0 \ k=0
m,
o/3m^.
/ mLOmu
2
}t
= e " * + e ^ * + 2e(" ^
0
2
(12-)
OO
= y >
^<
n=0
+ 2 L n ) ^
m n!
mn
The r e a d e r should now experiment with other simple functions involving powers of e.
A suggestion is to use some combinations which lead to hyperbolic sines or cosines, which
a r e defined in t e r m s of e.
5. GENERATING FUNCTIONS FOR MORE GENERALIZED IDENTITIES
To get identities of the type
n
F
= F
k+r
k=0
2n+r
Hi
n=0
so that if A(t) = (eat + ept)/(a
_ ^ / JL^ / n \
n=0 \ k=0
B(t) = e t ,
- p)t
tn
2-/ ( Z ) ( k ) Fk+r I ET
fn
n+r n!
= e D
/ e <*t _
/3t \
t ( 6 a -J
X
(13)
r "2t
a e
are(a+l)t
)=
^(p+Dt
^ I i3
'
RT
- P
a _ j3
Ph
= V*
Z-f
n=0
L
2n+r n!
1973]
281
n=0 \ k = 0
4m,
/D^m,
l
- eP
)/(a
,n
I - =
4mk+r I n !
4rm (a4m+l)t
(/3 4m +l)t
04rm
- ]3
e^
a - j3
4rm a
= 2
2
(14)
,
B(t) = el ,
- 0),
2m, 2 m , 0 2 m u
.
2m, 2 m I / D 2 m u
(a -f/3
)t
(a +/3
)t
p4rm a
e
a "- P
p
J
x
2m
2mn+4mr
n!
n=0
E^
,n
m,
t
t_ _ e
mn n! ~
n=0
(15)
Dm,
(aF +F
n )t
v
m m-1'
- er
_ e
a - j3
, , <*F t
j3F t \
(j3F +F n )t
^ m m-1'
- e
a - j3
/
n=0 \ k=0
Taking the r
oo /
=
(16)
JLJ F m n + r m iTT
n=0
2 ^ I 2 - M k I F m - 1 F m F k + r m J nf
n=0\k=0
/
Replace r m by
REFERENCES
1. V. E. Hoggatt, J r . , and D. A. Lind, n A P r i m e r for the Fibonacci Numbers: P a r t V I , "
Fibonacci Quarterly, 5 (1967), pp. 445-460.
2. H. W. Gould, "Generating Functions for Products of Powers of Fibonacci N u m b e r s , "
Fibonacci Quarterly, 1 (1963), No. 2, pp. 1-16.
3.
4. John Vinson, n The Relation of the Period Modulo to the Rank of Apparition of m
in the
n ^ 2,
and
1 k n - 1.
that
(0.
GCD
For n ^ 2, and l ^ k ^ n - 1 ,
we have that
((S:!).(r4(;f c^vM^M)
Proof.
G2,
respectively.
a + b
b + c
a + 2b + c
where
b+c = (M,
b + 2c + d).
etc.
Then
d
c + d
b + 2c + d
Gt = GCD(b, c + d, a + 2b + c) and
G2 = GCD(c, a + b,.
holds in these c a s e s , but one can see that Gt - G2 = 1 = GCD (a, b , c, d) directly.)
e I
el
e
We shall show that p | Gt if and only if p | G2, for any prime power p .
*This work was supported by a r e s e a r c h fellowship of the Italian National Research Council.
282
Oct. 1973
283
j c + d and p
p | a, p
|b,p
Lemma.
Suppose that p e I G ^
Then we have
| a + c, whence p | a and p j[ d.
|c
and p
| c + d a r e inconsistent.
|(
i)
implies p | k + 1.
Proof.
Let n = 2 a , p
and b = 2fo.p
p*i (:)
if and only if a is the number of borrows in the p - a r y subtraction n - k. Consider now b 0
and a 0 .
If
number of borrows.
If b 0 = a0
<
p - 1,
(nk)
|(
i)
pe|fk)
and
pe|(k + i)
implies p | n - k.
Proof. Use I\ ktx / I = I\ n "- .k /I and the Lemma
"'- Theorem,
^heo'rem, we
Returning to the
we have
e
P
hence p
= b
n - k + 1, and we have
hence p | k.
Thus p | n + 1. Now c + d = I
the p - a r y expansions.
1.
|(l:0"' * p, l(' : 0
k
1. Let n = ^ a . p 1 and k + 1 = s b . p 1 be
sult and
pe
e
F r o m Glaisher ! s r e -
H )-'
| c,
Gt = G2 = GCD(a,b,c,d).
284
Gt I d,
Oct. 1973
Conversely,
REFERENCES
1. H. W. Gould, "A New Greatest Common Divisor Property of the Binomial Coefficients,"
Notices Amer. Math. Soc. , 19 (1972) A-685, Abstract 72T-A248.
2. D. Singmaster, Divisibility of Binomial and Multinomial Coefficients by P r i m e s and P r i m e
P o w e r s , to appear.
Dear Editors:
On page 165 of Professor Coxeter T s Introduction to Geometry (New York, 1961), we
read: "In 1202, Leonardo of P i s a , nicknamed Fibonacci ("son of good nature"), came a c r o s s
his celebrated sequence . '
This translation of Leonardo's nickname differs, of c o u r s e , from the one ITve saen in
the Quarterly.
Who can solve the historic mystery for u s ?
Les Lange
Dean, School of Science
San Jose State University
San J o s e , California
Dear Editors:
Thank you for the reprints I have just received.
how the main sentence from "An Old Fibonacci Formula and Stopping R u l e s , " (Vol. 10, No.
6) was omitted.
The formula is
00
F(n)
9n+l
0
and it is based on Wald T s proof that the defined stopping rule is a real stopping rule (the p r o c e s s terminates after a final number of steps with probability 1).
R. Peleg
J e r u s a l e m , Israel
1.
INTRODUCTION
It is known that reflections of light rays within two glass plates can be expressed in
t e r m s of the Fibonacci numbers as mentioned by Moser [ l ] .
happens when the number of glass plates is increased.
tinct light paths, where n is the number of reflections and k the number of plates. Figure
1 illustrates the particular case of two glass plates for n = 0, 1, 2, and 3, where we a l ready know that the possible light paths result in the Fibonacci numbers.
upper surface in this figure indicate the s t a r t of a light ray for a distinct possible path for
each particular number of reflections.
I\i ^vi
P(0,2)=l
P(l,2,)=2
P(2,2)=3
P(3,2)=5
Figure 1
We will now derive a matrix equation which relates the number of distinct reflected
paths to the number of reflections and to the number of glass plates and examine a sequence
of polynomials arising from the characteristic equations of these m a t r i c e s .
285
286
[Oct.
reflection.
i,
Let Q(n,i)
l < i < k,
be the
at which
of k plates after exactly n reflections is identical to the number of possible distinct light
paths for n reflections,
(1)
P(n,k) = Q(n,k)
for all
and
k .
n+1 r e f l e c t i o n s
Q(n,0)
Q (n+l,k)
Q (n, 1)
Q (n+1,k-1)
Q (n, 2)
Q (n+l,k-2)
Q (n, k-1)
Q (n+1, 1)
Q (n, k)
Q (n+1, 0)
n reflections
Figure 2
11
ljl ||Q(n,k)
Q(n + 1, k - 1)
Q(n, k - 1 )
Q(n, 2)
oil IIQ(n,
Q(n + 1, k)
(2)
Q(n + 1, 2)
Q(n + 1, 1)
and define Q,
|l
1)
the minor diagonal all ones and with all zeros below the minor diagonal^
1973]
(3)
287
Q,,
1
kXk
Next, we find a recursion relation for the characteristic polynomials for the m a t r i c e s
Qk.
We let
Dk(y) = det(Q k -
(4)
where Q,
i s the identity m a t r i x of o r d e r k.
|i - y
I
ii
i - y
1 i
o
o
I
(5)
Djy) =
-y
-y|
kXk
The determinant on the right side of (5) is now modified by subtracting row 2 from row 1,
after which column 2 is subtracted, resulting in (6):
(6)
-2y
i - y
I
i - y
-y
-y k X k
This determinant (6) is then expanded by the elements in the first column, giving
(7)
by the elements in their l a s t row and column, respectively, they become At = -yD. Q(y) and
k_
A 3 = (-1) JD, 2 (y). If the determinant A2 is expanded according to the elements of its first
row, the resulting determinant according to the elements of its last row, and finally this new
determinant according to the elements of its l a s t row, one finds A2 = y 3 ^^ ^(y)* The above
expressions for A l 9 A 2 , and A3 a r e then substituted into Eq. (7), which yields the result
288
[Oct.
(8)
the desired recursion formulas for the polynomials D, (y) for all k ^ 1.
3. THE COEFFICIENTS OF D (y) AND RECURSION FORMULAS FOR P(n,k)
The polynomials D (y) can be expanded in power s e r i e s in y as
(9)
n-i
A
n^=S A n,iy
y
n,i
i=0
where A
and equating coefficients of like powers of y, the following recursion relation among the c o efficients are obtained:
(10)
where we take A
. = 2A 0 . - A 0 . 0 - A
..,
n,i
n-2,i
n-2,i-2
n-4,i '
0 <
. = 0 whenever
n = 2, 4, 6 and n = 1, 3, 5,
i < 0 or
n < i.
For
i < n ,
'
n, l
2n,0 "
A 0 ,- - = n + l
2n+l, 1
A 0 n = -n
2n,l
2n,2n
2n,2n-1
2n,2n-2 = ^
2n,2n-3 = ^
_1
2n+1,0
^
'
2n+l,2n
2n+l,2n+l
"
2n+l,2n-l =
"
~ 2>
2n+l,2n-2 = ^
- y( D n + 1
~ '
^
"
These sets of formulas for the coefficients will then permit one to write the polynomials
D (y) as power s e r i e s in y, which is very useful in obtaining recursion formulas for P(n,k).
1973]
289
that
(11)
^ A
P ( n - i , k) = 0
i=0
for all k.
Then, Eq. (11) is the recursion relation for the numbers P(n,k), and for k ^ 5,
we can write
P(n + 1, 1) = P ( n , l )
P(n + 2, 2) = P(n + 1, 2) + P(n,2)
P(n + 3, 3) = 2P(n + 2, 3) + P(n + 1, 3) - P(n,3)
P(n + 4, 4) = 2P(n + 3, 4) + 3P(n + 2, 4) - P(n + 1, 4) - P(n,4)
P(n + 5, 5) = 3P(n + 4, 5) + 3P(n + 3, 5) - 4P(n + 2, 5) - P(n + 1, 5) + P(n, 5) .
4. A GENERATING FUNCTION FOR THE POLYNOMIALS D (y)
Theorem 1. A generating function for D
(y) is
oo
(12)
Proof.
(13)
-1
= T>2n(j)t'1
n=0
'
In Eq. (12), multiply both sides by the denominator on the left side to obtain
[ D 2n + 4^ "
(2y2
"
1)D
2n+2(y)
^ n ^
Equating like powers of t on the left and right sides of Eq. (13), one obtains
D0(y) = 1
D2(y) = (2y2 - l)D0(y) - y(y + 1) = y 2 - y - 1
D 2 n + 4 (y) = (2y* - DD 2 n + 2 (y) -
^ ( y )
290
[Oct.
Since these l a s t three equations a r e in full agreement with the recursion formulas derived
e a r l i e r , one concludes that the theorem has been proved.
Theorem 2. A generating function for D
_ (y) is
OO
(14)
The proof of this theorem readily follows if one uses the same procedure as in proving the
preceding theorem.
The generating functions of Eqs. (12) and (14) can be used to obtain closed form solutions for D
(y) and D
(y) with the aid of the following equations from Rainville [2]:
(i - Z)"a = E
(i5)
n=0
(16)
n
(ft)n = T T ( a + k - 1) = a(a + l)(a + 2) (a + n - 1),
k=l
(a) 0 = 1,
a 0
n 2= 1 ,
.
<*>
<*> [n/2]
(17)
n=0 k=0 ^
'
n=0 k=0
'
n=0
(18)
= >J > - ( - i r K l ( 2 y 2 - i ) n - V k t n + k
n=0 k=0
[n/2]
E E
n=0
Hence,
k=0
291
y*t2f
2n=l k=0
^ k (\ n
(19)
n=l
k=0
k \ in,
^ n - 2 k 4k ,n , .,
] (2y - 1)
y t + 1
/
<*>
[(n-l)/2]
& DS
n=l
k=0
>
Now
2 D2n(y)tn = E D 2 n ( ^
n=0
n=l
(20)
2n ^
[n/2]
k=0
^h^^'^'^y^
^
(21)
[(n-D/2]
V*
/ i \ k / n - k - l \ / r > y 2 x-)n - 2 k - l
2^
^
(
k
f
"
k=o
\
/
/ , i\
y(y + D
2n,l^
= ^
(22)
E
k=0
[(n-l)/2]
-y
E
k=0
4k
'
[n/2]
D
'
/
^
2
^
^n-2ky4k
'
,
'
REFERENCES
1.
L. Moser and M. Wyman, Problem B-6, Fibonacci Quarterly, Vol. 1, No. 1, Feb. 1963,
p. 74.
2.
3.
Bjarne Junge, "A Study of Matrix Developed Sequences and Polynomials, n M a s t e r ' s Thesis, San Jose State College, San J o s e , Calif. , June, 1963.
In the present paper, it is shown that the main theorem in [ 1 ] , see p. 484, can be e s tablished by using one of J. G. van d e r Corput ? s difference theorems [2].
Moreover, by
using a theorem of C. L VandenEynden [3] we show the property that the sequence of the
integral p a r t s of the logarithms of the recursive sequence under consideration is also uniformly distributed modulo m for any integer m > 2.
Lemma 1. Let (x ), n = 1, 2, , be a sequence of real numbers.
If
lim (x , - - x ) = a ,
n oo n+1
n
'
a irrational, then (x ) is u. d. mod 1 ( [ 2 ] , p. 378).
Lemma 2.
Assume that
m > 2.
Then the
[3] .
n >
1 ,
'
Vt = n .
V2 = y 2 ,
k-1
aj i x
_ . . _ a-jx an
Vk = yk
Then:
Proof.
is u. d.
Oct. 1973
293
3=1
where the coefficients av a2i , a^ a r e uniquely determined by assumption (2).
be the l a r g e s t value of j for which a. f- 0. We have p > 1.
Let p
Hence
Vn = ZYJ a.] $] .
K
j=l
Now
,7
V ,n
n+1
^n+1 .
,
/?n+l
i^i
+ + ct P
iri
p p
5;
^Ha
-* ^D
QfijSf + . . . + a f
1 1
p p
as
~*
'
since jSj | jS 0 (as noo), i = 1, 2, , p - 1, because of the conditions on the a b s o lute values of the p..
The number j3
g v n + 1 - l o g V n - log/3 ,
as
(log V ) is u. d. mod 1.
This
Hence
proves
logV
p
*p
-* _x.
as
n oo ,
m
m
m
hence the sequence ((log V ) / m ) , n = 1, 2, is u. d. mod 1, and according to Lemma
&
([log V ] ) is u. d. mod m
for
every
integer m > 2.
Remark. By restricting the o r d e r of the r e c u r r e n c e we may relax the conditions on the
coefficients
(V ) can be nega-
tive in that c a s e , and so we obtain a result regarding the logarithms of the absolute value cf
V
Let (V ), n = 1, 2, , be a sequence generated by the recurrence
n+2
where Vj = y 1} V2 = y2Yi and y2
2
are
^ J
and
We
fy
1 n+1
0 n
i not both 0.
where
nomial x - aAx - a0 has distinct real roots, jS1 and j32, one of which has an absolute value
294
Oct. 1973
inte-
We have
Vn
lih^T1
,
where
Pi
Hai
^ l
4a0),
j32 = \(^
- ^
+ 4a 0 )
Now
log | v n + 1 | - l o g | v n | = log
n+1
(72 - 7ihWi
- (72 -
(72 - YiMPT1
7iiW
- (72 - 7ii3i)i3 2 1 - 1
< 1.
algebraic, we may complete the proof in the same way as done above.
REFERENCES
1. J. L Brown and R. L. Duncan, ''Modulo One Uniform Distribution of the Sequence of
Logarithms of Certain Recursive Sequences," Fibonacci Quarterly, Vol. 8, No. 5(1970),
pp. 482, etc.
2.
3.
For
+ u - v\fS)
read
For
For
For
For
r2^
rc:
For
i. e.
F o r N = orthocenter read
For
|l = H | 2
read
|l - H | 2 .
residue
read
radius .
K2
a2
read
4x 2 - 3y2
, 6 , 14
Underline
+ (u - W*3).
s2
- 3v2
read
4X 2
read
r2
read
r c :<*>, 6,14
read
as
H = orthocenter.
rather than
Dreieck.
1.
INTRODUCTION
In a previous paper [l] the idea of a "sieve" was extended in o r d e r to give a method for
the computation of the sequences of values for certain functions which occur in the theory of
numbers.
suitable Dirichlet s e r i e s or of suitable power series; see, for example, G. H. Hardy and
E. M. Wright [2, Chapters 17 and 19], We will consider the similar problems of the number of representations of an integer as a product with factors chosen from a given set of p o s itive integers and the problem of the number of representations of an integer as a sum with
t e r m s chosen from a given set of positive integers. Although quite analogous, the two p r o b l e m s are r a r e l y mentioned together.
To be specific, we consider a subsequence
infinite, which satisfies the conditions aA < a2
ucts and 0 < aA for the case of sums.
number,
S = {a },
<
a3
<
with a. G S, b.
>
P(n),
of distinct representations of a
that i s , of
distinct factors, and (2 ) compute the number, P (n), of distinct representations of a number
n as a sum of the form n = a. + a. + , a. E S, that i s , of distinct t e r m s . The generatj
and P ! a r e given by
11(1 - a"8)""*
11(1 - x V
= Rfci)n~ B
=
X>Cn)x n ,
0 ( 1 + a"8)
= ER'(n)n'S
11(1 + x8"1)
= X>!(n)xn
295
296
[Oct.
The reciprocals of these generating functions a r e also of interest. F o r example, those for R
and P lead to generalizations of the Mobius function (ft(n) is obtained for a. = p.) and of the
Euler identity (whis is obtained for a. = i),
respectively,
11(1 - a7 S ) =
X)M(n)n - s
0 ( 1 - x *) =
K(n)xn
2. REPRESENTATIONS AS A PRODUCT
F r o m the generating function we can develop a "sieve" technique ("sieve" used in the
sense given in [l]) for computing the values of the sequence R.
nd
a-s)_
n
IH1
a-
a;2s
) ,
\ a } f o r l n N
R(1)
R(2)
R(3)
R(4)
1
2
3
5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2
1 0 0 0 0 0 0 0 0 0 0' 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 1 0 1 0 0 0 1 0 0 0' 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
1 1 1 1 0 1 0 1 1 0 0 1 0 0 0 1 0 1 0 0 0 0 0 1 0 0 1 0 0 0 0 1
1 1 1 1 1 1 0 1 1 1 0 1 0 0 1 1 0 1 0 1 0 0 0 1 1 0 1 0 0 1 0 1
1 1 1 1 1 1 0 2 1 1 0 1 0 0 1 2 0 1 0 1 0 0 0 2 1 0 1 0 0 1 0 2
1 1 1 1 1 1 0 2 1 1 0 1 1 0 1 2 0 1 0 1 1 0 0 2 1 1 1 0 0 1 0 2
To begin, let R
2 n < N. In o r d e r to obtain the sequence R,(D we bring down 1 into position 1 and then
add the entry from position 1 of R
sum in position aA of R
la-j of R
in position 2aA of R* ' and the sum is entered in position 2a! of R ' ' ,
n ^ ma! of RW we simply use the entry from position n of R
ses using a^ is stopped when maj
entered in position 1 of R
successively for 1
<
>
etc.
At
position
is computed
1973]
n of R
if n j4 ma 2 ,
297
2
of R^ '
to the
entry at position ma 2 of R ' ' and entering the sum in position n of R . F o r the general
to R^ k ',
since
R^(n)
R ( k ) (n) = R ( k - 1 } (n)
if
aR j n ,
R ( k ) (n) = R ^ " 1 ^ ) + R ( k ) ( n / a k )
if
ak | n .
We stop each subprocess when ma^ > N and we stop the entire process when a^ > N;
the
result is the sequence of values of R for 1 < n < N. It should be noted that the sequence
p(R-i;
can
function
k-1
_i
H (1 - a " s )
.,
n
n=i
by the s e r i e s
(1 - a " 8 ) " 1 = 1
a" s
a"2s
equivalent to successively expanding the scale for Rlk-D by a factor a k and adding the expanded result to the sequence R^
is denoted by c
n
Table 2
Product Pattern
ClC2
c
m
'
ak'
Ci
2 a ^ e''
c
''
"*
a k '" '
mak2
ma]j
F r o m the diagram in Table 2 we note that rows, beginning with the second, are repeats
of the first row, but with the scale expanded successively by the factor a^. If we consider a
column which is to be summed, we note that the quantity to be added to c n a
is m e r e l y the
sum appearing in the column headed by c , hence we obtain the iteration equations.
In the intermediate steps the number R^'(n) has an interpretation as the number of
representations of n as a product using only elements of the finite subsequence a l s a 2 , ,
298
[Oct.
Here we have a quite closely analogous case to that of the previous section.
The gen-
n(i - x ) n(i + x n + x
n
2a
+ ...) .
In Table 3 the operations a r e diagrammed analogous to Table 2 except that here we notice that
the rows a r e shifted by an amount a.
(k - 1)-sequence.
Table 3
Sum Pattern
C 0 Ci C 2 ' C
' akCak+1"'C2ak"-C3^'
c
c m a k
* c a k ' c 2a k c(m-l)ak
o X - (m-2)a
c
co * "
(Here the sequence is indexed from 0.) Reading the columns as in Table 2 we have the interation formulas
p(k)(n)
p(k-l)(n)
if
| n ,
aR < n .
1, 2, 3, 5, 8,
if
is illustrated.
Sum Representations,
Table 4
S = { l , 2, 3, '5, 8, 13, }
0 1 2 3 4 5 6
9 10 11 12 13 14 15 16 17 18 19 20
21
yoT
1 0 0 0 0 0 0
P(i)
1 1 1 1 1 1 1
p(2)
1 1 2 2 3 3 4
9 10 10 11
11
p(3)
1 1 2 3 4 5 7
8 10 12 14 16 19 21 24 27 30 33 37 40 44
48
p(4)
1 1 2 3 4 6 8 10 13 16 20 24 29 34 40 47 54 62 71 80 91 102
1973]
299
If it is required that the representations involve no repeated elements, that i s , " s q u a r e free" products or "pairfree" s u m s , the generating functions a r e simpler and hence the computational process is also simpler.
+ R'^Wak)
if
a, \ n ,
if
| n ,
where we s t a r t with R f ^ ( l ) = 1 and Rf^ '(n) = 0 for n > 1. The case of distinct t e r m s
Starting with P ! ^ J>(0) = 1, P 1 ^ '(n) = 0 for n > 0 we
P . ^ (n)
if
a, * n .
if
a,
p i * - ) ^ ) + p i ^ - ^ d i - ak)
< n
The only alteration required to change the formulas of the previous sections to these
is the change in the upper index on the second term. Tables 5 and 6 display the computations
of Rf and P f for the Fibonacci sequence.
Table 5
Squarefree Products, S = {2, 3 , 5, 8, 13, }
1
1 JL J3_J^__J3_LL JL_9_JL JLJLJ5_JL A
JLJ7_JLJLJ)__1_J_j^_4_JL_6_J7__8_J)_J)_J^ 2
H.(0)
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Rf(D
1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
R.<
1 1 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Rf (3)
1 1 1 0 1 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0
1 1 1 0 1 1 0 1 0 1 0 0 1 0 1 1 0 0 0 0 1 0 0 1 0 1 0 0 0 1 0 0
Pairfree Sums,
0 1 2 3 4 5 6 7 8 9
1
0
Table 6
S = { l , 2, 3, 5, 8, 13, - }
2
3
1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
P.C0)
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
P ,u>
1 1 0 0 0 0 0
0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
(Continued on following page.)
300
[Oct.
3
9 0 1
p.<2>
1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
P.(3)
1 1 1 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
p. <4
1 1 1 2 1 2 2 1 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
p.(5>
1 1 1 2 1 2 2 1 3 2 2 3 1 2 2 1 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0
1 1 1 2 1 2 2 1 3 2 2 3 1 3 3 2 4 2 3 3 1 4 3 3 5 2 4 4 2 5 3 3
5. OTHER F O R M S
As was remarked in the first section, the reciprocals of the generating functions a r e
also of interest.
Table 7
Product "Reciprocals" S = -j 2, 3, 5, 8, 13, }
(0)
T (wD
M
M
(2)
(3)
3 4
6 7
8 9
1
0 1 2
3 4
6 7 8 9 0
+1
0 0
0 0
0 0
0 0 0
0 0
0 0 0 0 0
+1-1
0 0
0 0
0 0
0 0 0
0 0
0 0 0 0 0
0 + 1 0
0 0
0 0 0
0 0
0 0 0 0 0
0 0 0 0 0
+1-1-10
+ 1 - 1 - 1 0 - 1 + 1 0
0 0 + 1 0 0
0 0 + 1
- 1 0 - 1 + 1 0 - 1 0 + 1 0 0 - 1 0 + 1 + 1 0 0 0 0
+1
Table 8
Sum " R e c i p r o c a l , " S = { l , 2, 3, 5 , 8, 13, }
K (o)
K(l)
K (2)
K (3)
K (4)
1
0
4 5 6 7
2
0 1
+1
0 0 0 0
0 0
+1 -1
0 0 0 0
0 0
+1 -1 -1 +1
0 0 0 0
0 0
+1 -1 -1
0 +1 +1 -1
0 0 0 0
0 0
+1 -1 -1
0 +1
0 +1
0 -1 -1 +1
0 0 0 0
0 0
+1 -1 -1
0 +1
0 +1 -1
0 +1 -1 -1 +1 0
0 0 +1 -1 -1 0
19
?3]
301
Note that the only alterations required in the techniques used involves the changes of signs;
this is equivalent to switching the primed and unprimed formulas and subtracting the second
t e r m s instead of adding these t e r m s .
All of the cases which have been considered are special cases of the general formulas
n ( l + f(ak)a^s + f(a^)a"2s + . . . )
a
n ( l + g(a k )x
=J>(n)n~s ,
2a,
+ g(a^)x ^ + ) = G ( n ) x n .
Other cases can certainly be derived and similar lines of reasoning can be c a r r i e d out for the
simpler c a s e s .
REFERENCES
1. R. B. Buschman, "Some Simple S i e v e s , " Fibonacci Quarterly, Vol. 1 1 , No. 3 S pp. 247 254.
2.
ERRATA
Please make the following changes in the article, "A New Look at Fibonacci Generalizations, " b y N. T. Gridgeman, appearing in Vol. 11, No. 1, pp. 40-55.
Page 40, Eqs. (1) and (2). Please insert an opening bracket immediately following the
summation sign, and a closing bracket immediately following " B " in both c a s e s . In Eq. (2),
please change the lower limit of the summation to read: "i=0" instead of "m=0. "
Page 4 1 , Table 1. Please add continue signs, i. e. ,
'.
'.
Page 42, line 14 from bottom: Please c o r r e c t spelling from " s u p e r f l u o u s " to " s u p e r fluous. "
Page 42, line 7 from bottom: Please insert a space between "over" and "positive. "
Page 44, line 10: Please change " m e m b e r "
to read
"members."
Page 45, Eq. (11): Please add an opening bracket immediately before I
__ 1 1 and a
MULTIPLE REFLECTIONS
LEO MOSER and MAX WYMAN
University of Alberta, Edmonton, Alberta, Canada
(glass)
(mirror)
Let f (n) be the number of paths which s t a r t at the top of the plate and reach plate r
after n upward reflections.
F u r t h e r , let A = (aA-) be a k X k
that
(1.1)
f (n)
rj f j ( n "
1]
3=1
If F(n) denotes the one column matrix
fi(n)
F(n)
(1.2)
f2(n)
fk(n)
(1.3)
Hence, by iteration we have
(1.4)
F(n) = A1 F(0) .
Thus (1.4) provides an explicit solution for F(n) in t e r m s of F(0).
suitable to compute the asymptotic behavior of F(n) for large values of n. We now d e r i v e s
second explicit form by means of which the asymptotic behavior is easily calculated.
302
Oct. 1973
MULTIPLE REFLECTIONS
303
(1.5)
A1*"1 + . . . + c 0 = 0 .
and
Let us assume that the roots of (1.5) a r e Als A2> " *" A.
tnat
tne
s e roots a r e distinct.
The case of multiple roots can also be treated easily by the method we shall use.
Since every matrix A satisfies its own characteristic equation we also have
(1.6)
Multiplying by A
n-k
A k + c.
A ^ 1 + . . . + CQI = 0 .
, we have
A n + C j ^ A 1 1 - 1 + . . . + c 0 A n " k = 0.
(1.7)
F(n) + c k _ 1 F ( Q - 1) + . . .
+ C 0 F ( Q - k)
= 0 ,
and
(1.9)
f r (n) + c k __ 1 f r (n _ 1) + . . . + c 0 f r (n - k) = 0 .
(1.10)
'"
+ C
<lai>
f (n)
B X
rj jj
'
3=1
where the constants B, . do not depend on n.
the m a t r i c e s 1, A, , A
B ^ B ^ , , Brk.
Explicit expressions
for
these constants can be given. From (1.10)we can easily seethe asymptotic behavior of f r (n).
Let us write A. in the form A = r. exp (id.).
> r p + 1 > r p + 2 > . . . > rk.
Clearly,
k
(1.12)
f r fe) -
rf 2
By exp (in0 ) .
3=1
If p = 1 then
(1.13)
f (n) -
A? B T
rj = r 2 = = r
304
[Oct.
MULTIPLE REFLECTIONS
Example. If the matrix A is given by
(1.14)
A =
(A) =
| x i .- A l
(1.16)
= 1 -A
Dk(A) = H
,
vk
(1 - 2A) + ^(1 - 2A)2 - 4A2 \
(1.17)
+ K
)'
HRj + KR2
where H, K a r e constants depending on A but not on k.
we find
(1.18)
2(1 + cos 6)
cos k.6 +
1 + cos
sin i
SGC
sin k# = 0 ,
sin (k +\)Q
:n
cos % 6 = 0
sin v
*
Obviously, the roots of (1.19) are
(1.20)
6 =
sir
k +
s = 1, 2,
1973]
where
MULTIPLE REFLECTIONS
305
s = 0 m u s t be e x c l u d e d b e c a u s e of t h e d e n o m i n a t o r .
H e n c e t h e r o o t s of D, (A) = 0
a r e given by
(1.21)
A = { sec* I ^ r
1>
-\*(W$T)>
s = ls 2,
(1.22)
At
Thus
(1.23)
f (n)
rv '
-^^(ffiVrJj1
rl
TWO N U M E R I C A L CASES
C a s e 1. k = 2 .
1 -A
1
D2(A)
1
2 -A
.
_ 3 + ^ 5
At _
s
9
fi(0)
= 0,
f 2 (0) = 1 ,
fj(n)
_ 3 - N/5
A2 ~
o
fj(l) = 1 ,
= Buxf
0 =
A^ - 3 A + 1 = 0 ,
f2(l) = 2 ,
+ B12An
+ B12
1 = B ^ A i + B12A2
Bu
Ai - A2
and
B12
Ai - A2
t 1 \
1 / A n x n \ _ 1 ^n (1 ^2n^
fA(n) = (At - A2 ) = Ai (1 A2 )
1 / s + ^Y 1
Similarly,
f 2 (n)
= B 2 iAi + B 2 2 A 2
1 - B2i + B22
2 = B 2 iAi + B 2 2 A 2
2
B
21
2 _ ! + ^5
Ai - A2
2^5
2 - Ai
22
A2 - Ai
1 . ^5
2\/5
306
MULTIPLE REFLECTIONS
Oct. 1973
Xf = I s e c ^ f ) ,
X2 = i s e c ' ( ^ j ,
A3 = {
t(0) = 0
f2(0) = 0
f3(0)
= 1
fjd)
f2(l) = 2
f s (l)
= 3
f2(2) = 11
f3(2)
= 14
= 1
ft(2) = 6
s e M \
Thus
f3(n) = B 3 1 xf + B^A? + BS3xf
5
3i
3 - B^iXi
14 = &$\\t
"* Boo
^32
+
"*"Bo;-^33
+ B32X2 + B33A3
+ B32A2
B33A3
Solving simultaneously,
A2A3 - 3(A2 + A3) + 14
B 31 =
(n) f8(n
so that
2n
0.537 U sec ( H J V
Page 49, Eq. (33): Please change the last number on the line from " 3 " to " 1 . M
Page 49, Line following Eq. (34): Please raise
" (mod 3)
the
identity
p(n) = p(n - 1) + p(n - 2) - p(n - 5) - p(n - 7)
+ p(n - 12) + ptn - 15) - - + + . . .
(1)
=^(-D
i + 1
- ^ (3i2 + i)) ,
p(n
(2)
where the sum extends over all integers i for which the arguments of the partition function
a r e nonnegative.
Proof.
oo
q(n)x
n=0
We have
oo
^ ( 1 + x
=JJ(l-xJ)"
i=0
j=0
)xj
(i (x2)i>
\j=0
i=0
3i +i
j=0
\ i=-oc
/
307
j=0
308
and the
e result follows by equating <
coefficients of x
Corollary.
[Oct.
If n ^ 0 , then
q(n) = p(n) + ^ ( - 1 ) ^
i=l
Theorem 2. If m ^ 3 , 1 -* a < m / 2 , n > 0 and q(a,m,n) is the number of p a r t i tions of n into distinct parts congruent to a (mod m ) , then
(3)
pf^-^lf1)
q(a,m,n) =
L|(n+aj)
Proof. We have
, n
n=0 q(a,m,n)x
I Ir,- ,
im+aw- ,
= i=0
| |((1+ x
im+m-a v \
)(1 + x
)}
i=0
n
oo
oo
/im+m " T T / / i
i m + m W l , i m + a W l , im+m-av\
(1 - x
)
{(1 - x
)(1 + x
)(1 + x
)>
i=0
i=0
\ -A
rm
/i +
, xn n + a - m W)(1
r m - a ))
^ .
l _
= II \>~p^( i/-\
) x im IJ TI TM/ /(11 - x \)(1
+, x
J=0
r=l
By Jacobi T s identity,
oo
oo
2r
IT
q = x
we find
and
z = x
"T
1973]
- xrm)(l
f]{(l
xrm+a-m)(l
xrm-a)}
= J2
r=l
1 1 1
^ " ^
309
j=^o
Therefore,
n=0
\ i=0
/ \j=-oo
2^q(a,m,n)xn
p(i)xiXXX J J J^M x
= I ^
n=0
\i=-o
\j=_.
Thus,
q(a,m,n)
=/])
n -(m(iJj^) - aj]\
where the sum extends over all integral values of j for which
n - y m (ilJ J . aj J
m
is an integer.
q(a9m n) =
Therefore,
2 p(!H^-j-fj)
'
m|(n+aj)
as required.
Corollary.
a
If (a,m) I n,
' = T575)
mi =
Let
Taflj
aij =
then
and n ^ 0.
-J5JE)
(mod mi)
'
310
jjj + j 0 \
(m\
2~)
\~
\3mi
l I~m
mx
+
[Oct.
* y
T - ai / k i
(n + aj),
There-
aij 5
q(a,m,n)
/ n + a(j 0 + kmj)
L H-"^
k=-oo
*
By Theorem 2,
p(-Lai_ji_U)
^
'
jEJoCmodmi)
(j 0 + k m ^ 2 + (j 0 + knij) \
'
/ / n + aj 0
]J + j 0 \
/ m?
/
Jom!
m.!
anij
- _
k]
But
* _ _a_ m _
a
m (a,m) (a,m)
ai
q(l,3,n)
+ k
+P
1973]
311
2^-fkl
q(l,3,n)
l*+i^
+ p ( ^ - - )
P ( ^ - 3 3 )
P ( ^ -
8 )
q(l,3,n)
J>(->^))
(n
0O
k=l
(H1-)*>>(JIT1-)+>'(H2-)REFERENCE
1. Ivan Niven and Herbert S. Zuckerman, An Introduction to the Theory of Numbers, 3rded.
John Wiley and Sons, Inc. , New York, 1972, pp. 226-227.
results.
=0
(mod p)
(
F
m+1 "
(mod
p)
if m is odd, then
L
L
=0
m+1 "
(mod p)
v
(mod
p)
()
H-222
(G)-
-1)
ii
m
(
Proposed by Ft. E. Whitney, Lock Haven State College, Lock Haven, Pennsylvania.
A natural number,
r e p r e s e n t s the sum of the distinct divisors of n (not necessarily proper). Finally a number,
n, is called weird* if it is abundant and not semiperfect.
Are any Fibonacci or Lucas numbers weird?
etc.
A D V A N C E D PROBLEMS A N D SOLUTIONS
O c t 1973
H-224 Proposed by V.
313
Let
1
1
4
14
A = I 2
10
22
nXn
14
10
22
n + 1
nXn
where the first row and column of A have been deleted.
Show
(i) A = 1 and
(ii) B = n + 1.
H-225
x, y,
and z.
H-226
z - y
(a1?
(i = 1, 2, , n)
, a ) such that
n
0 < a. < k
(i = 1, 2, , n)
l '
l-l
314
Oct. 1973
SOLUTIONS
PROBABLY?
H-179
Set a
= 0 for
n < 0;
a0 = 1 and d e -
by the recursion
n
1.
n
=\
p. a .
n
Ci *i n-i
i=l
n > 0 .
lim
to exist
c) infinite.
>
and
i=l
sufficient for
lim
n -+00 n
other random device) such that the probability of i occurring is p.. We have a game board
consisting of a sequence of squares indexed 0, 1, 2, .
the die to determine the n number of squares moved, as in Monopoly or other board games.
Then a
is
n
= S * Pi '
i=l
If all p. = l / k ,
= 1, a 2
- = 0 for all n.
Editorial Comment
i > k,
(1)
where p is an odd prime number
>
1 9 n ^ 2 and 1 < xA ^ ^ x n 9
is known to p o s s e s s
general solutions for n = 2, p ~ 2; n = 3, p = 3; for other values of n and p , no general solutions a r e known9 although computer searches for solutions of such equations can
easily be c a r r i e d through by assigning a value to y and n and then allowing the corresponding x ! s to take all values from x t = ' = x n * 1 to xA = = x n = y - 1; in each c a s e ,
different p r i m e s p a r e tested to see whether Eq. (1) is satisfied.
ever, is drastically reduced by realizing that for a given n and y possible p r i m e s p which
can satisfy Eq. (1), have an upper bound, above which no solutions a r e possible. This s t a t e ment is a consequence of examining properties of the function \p which is defined by
(2)
i/r = (xj + . . . + x n ) / y
where y is given by Eq. (1)9 subject to the restrictions stated above. The relevant property
of ip is given by the following:
Theorem.
- 1^
depending on whether the solution to Eq. (1) are integers or not, respectively.
Proof of Theorem.
gg - 0
(1 i s n)
The conditions
| i = 0
and
| - 0
<*'
xf + *. + x? + . + x p - xf" 1 (xA + + x. + + x n )
315
= 0
or
316
Oct. 1973
v(4)
'
Subtracting Eq. (3) from Eq. (4) gives the condition for a turning point as
(5)
xx - x2 . . .
= xn .
1-1
F r o m which we deduce that max (t//) = n
. The lower bound of x depends on whether the
xs a r e restricted to integers. Thus if the x f s a r e non-integral, then we note that y p =
xf + + x p < (xi + + x n ) p so that 1 < i/>. If the x ' s a r e integers only, we use the
little F e r m a t theorem x P = x. (mod p). But since (xP - x.) is even and p is odd by h y pothesis, it follows that x p s x. (mod 2p) and hence using Eq. (1) we deduce that xt +
+ x m s y (mod 2p) from which it follows immediately that y + 2p < xt + + x . The
case of p ^ i deserves special attention. Using the same reasoning as above, we obtain the
inequality:
(6)
_
y + 2 < xx + + x . < \l n y
for any n by using the well known general solution for n = 2 i . e . , the identity
2
(a - b )
= (a + b ) . Thus putting a = n,
(2ab) 2 +
and b = n + 1, we obtain:
(7)
with U = m (m + 1). It is easy to use induction to show that this method gives an identity in
m.
b.
to obtain an identity in a
and
+2
= F
+F
for
n > 1.
?,
n > 0.
The theorem holds for n = 0, for when we have a set of no integers the only
F1+2 = F3 = 2
n = 2, 3 subsets: { l } , {2}, 0;
F2+2 = F4 = 3
F3+2 = F5 = 5 .
k+2
k+1
k+3
(k - 1)
F,
inte-
subsets
F. + 1
subsets
(k+l)+2
by mathematical induction.
Now, for the application.
or ones (every box containing exactly one of those numbers) such that no two "ones" appear in
317
318
adjacent boxes is F
2-
[Oct.
(To apply the lemma simply number the n boxes.) Since we do not
wish to use all zeros (<#>, the empty set in the lemma) the number of logically useable a r rangements is F
The interpretation to give the " z e r o " or "one" designation is whether or not
one uses that particular Fibonacci number in the given representation. If a one appears in the
box allocated to F, , then F,
no two adjacent boxes can each contain a " o n e , " no two consecutive Fibonacci numbers may
occur in the same representation.
Since the following a r e easily established identities,
2 +
4 +
...
+F2k
= F2fc+1-1
'2k+l
2k+2
using the Fibonacci positional notation the l a r g e s t number representable under the constraint
with our n boxes is F ^ 0 - 1. Also the number F ,., is in the n
box, so we must be
n+2
n+1
able to represent at most F ? - 1 distinct numbers with F 2 , F 3 , , F n + 1 subject to the
constraint that no two adjacent Fibonacci numbers are used.
Since there a r e F
- 1 different integers.
- 1 dif-
Fibonacci numbers if no two consecutive Fibonacci numbers are used in the representation.
We shall defer the proof of this until a l a t e r section. Now, a minimal representation of
an integer p uses the least possible number of Fibonacci numbers in the sum.
If both
and F,
thereby r e -
is
F,
a minimal
representation
and
Zeckendorf
representation.
2.
ENUMERATING POLYNOMIALS
Next, we use enumerating polynomials to establish the existence of at least one minimal
representation for each integer.
An enumerating polynomial counts the number of Fibonacci numbers n e c e s s a r y in the
representation of each integer p in a given interval F
Associated with this interval is a polynomial P
in the interval F
< p < F
..(x) if
-, there a r e a_ integers p whose minimal representation r e F o r example, consider the interval F 6 = 8 < p < 13 = F 7 .
1973]
319
8 = 8
9 = 1 + 8
10 = 2 + 8
11 = 3 + 8
12 = 1 + 3 + 8 .
Thus,
required 2 Fibonacci numbers, and one integer required one Fibonacci number in its minimal representation.
contain 8 but not 5. We now list the first nine enumerating polynomials.
F
m = 1
P < F x1
m
*
m+1
1 p < 1
^
Ax)
m-1
0 = P0(x)
m = 2
1 < p < 2
x = P4(x)
m = 3
2 < p < 3
x = P2(x)
m = 4
3 < p < 5
x2+x
= P3(x)
m = 5
5 < p < 8
2x2+x
= P4(x)
m = 6
8 < p < 13
x 3 + 3x2 + x = P5(x)
m = 7
13 < p < 21
21 ^. p < 34
34 < p < 55
x4 +
We shall now proceed by mathematical induction to derive a recurrence relation for the
enumerating polynomials P
nomial for F
and F
< p < F
< p <F
?.
+2
(x). It is evident from the definitions that an enumerating polyis the sum of the enumerating polynomials for F m < p < F m + 1
= T ? + T ?
m+1
< D < F
m+2 ~
+F
m+1
=F
m+2
Jx)
m+3
since each integer
p in
this interval has one more Fibonacci number in its minimal representation than did the c o r responding integer p in the interval F
< p < F
-.
< p
<
require an
- p < F
in this
have F
320
and put in an F
2-
[Oct.
The resulting integer will have a minimal representation with the same
number of Fibonacci numbers as was required before. In other words, the enumerating polymonial P
- F
+ F
m+2
m+1
+2
< D
m+1 ""
< p? < L
< p < F
< F
- and
+.
m+2
+F
< pT < F
=L
m+1
m+2
m+1
is
Pm+1W
= P m (x) + x P m _ 1 ( x ) ,
P0(x) = 0,
Pi(x) = x ,
< p < F
- contains F
but not F
steps. Each minimal representation in the interval F IO < pF < F IO contains F lrt ex*
^
m+2
m+3
m+2
plicitly since we added F
to a representation from the interval F < p < F
- and by
the inductive hypothesis those representations did not contain F
- but all contained F .
Next, for the representations froni F
^ < p < F ? , all of which used F
- explicitly
by inductive assumption, we removed the F
- and replaced it by F ? so that each r e p resentation in F ^ 0 < p < F 10 contains F J O but not F ( 1 . Thus, if the integers p
&
m+2 *
m+3
m+2
m+1
'
^
in the previous two intervals, namely, F 5 P < F
- and F
.. < p < F ? , had
Zeckendorf representations, then the representations of the integers p in the interval F
<
p < F
9.
But,
(1) = F
because PH(1)
= P 2 (l) = 1 and P ^ ( 1 ) = P (1) + 1 P
,(1), so that the
1W
m
m
*
m+1
m
m-1
two sequences have the same beginning values and the same recursion formula. The number
<
p < F
, is F , o - F .., = F , so that everyJ inteJ i
m+1
^
m+2
m+2
m+1
m
Thus, at least one minimal representation exists for each integer, and
of integers
in the interval F
&
ger is represented.
we have established Zeckendorf s theorem, that each integer has a unique minimal r e p r e s e n tation in Fibonacci numbers. Notice that this means that it is possible to express any integer
as a sum of distinct Fibonacci numbers. Also, notice that the coefficients of P
summands along the diagonals of P a s c a l ' s triangle summing to F
(x) a r e the
resentation.
Returning to our n non-empty boxes, for this case we wish to fill the boxes
1973]
321
between the left-most one and the next on the right, between any adjacent pair of ones, and
on the right of the last one if necessary.
2
3
2 n possibilities, o r ,
1, 2,
- l < p
<
F - l
an enumerating
maximal polynomial P * . (x) which has a term axJ if a of the integers p require j _ F i b onacci numbers in their maximal representation.
F6 - 1 = 7
Thus,
10
= 5 + 3 + 2 + 1
11
= 5 + 3 + 2 + 1 .
require 3 Fibonacci numbers, and one integer requires 2 Fibonacci numbers in its maximal
representation.
Notice that all maximal representations above use 5 but none use 8.
The
F - l < p < F
m
m = 2
m
- l
PL ,W
m+1
0 < p < 1
Hi-1
1 = P*(x)
1 < p < 2
x = P2*(x)
2
m = 4
2 < p < 4
x + x = P 3 (x)
m = 5
4 < p < 7
x 3 + 2x2 = P 4*,(x)
m = 6
7 < p < 12
x4 + 3X3 + x2 = P5*(x)
m = 7
12 < p < 20
x 5 + 4x 4 + 3x3 = Pg (x)
20 < p < 33
m = 9
33 < p < 54
As before, we now derive the recurrence relation for the polynomials P*(x).
Lemma.
F
F
, i - l contains explicitly
F
.
F
J
m+1
m - 1n
Proof. We can add F
to each maximal representation in the interval F
, -, - 1 and these numbers fall in the interval
m+1
2F
, - 1 .
m+2
- 1 <p <
322
[Oct.
, - 1 < pT < 2F - 1 .
m+1n
^
m
L 1
m+1
- l < p < F _ L O - l .
^
m+2
Pf(x) = 1,
P2*(x) = x ,
and again P*(l) = F . This establishes that each non-negative integer has at least one maximal representation.
Returning to the table of the first eight polynomials P*(x), by laws of polynomial addition, adding the enumerating maximal polynomials yields a count of how many numbers r e quire k Fibonacci numbers in their maximal representation.
tation in distinct Fibonacci numbers, which proves the Dual Zeckendorf Theorem [2]:
Theorem. Each positive integer has a unique representation as the sum of distinct F i b onacci numbers if no two consecutive Fibonacci numbers are omitted in the representation.
Lemma. Let fx(x) = 1, f2(x) = x , and f
polynomials. Then
+2(
n-l
= xf
+i( x )
n > 0.
Observe that
Pf(x2) = 1 = xHM ,
p*(x2)
p* (x 2)
x2 =
x 2[p*_ i ( x 2)
+ f
x lf 2 ( x ) f
P*_2(X2)] .
^ ^e
the
FiDOnacci
1973]
323
Assume that
P * o<x2) = x n - 3 f
Q(x)
Thus
P*(x 2 ) = ^ [ xn-2^ r ^ x -) . +, xnn--3" ,\ _ 2 ( x ) ]
+ f
= ^K-l^
Lemma.
n-2 ( x ) 3 =
xI1 lf (x)
" n
oo
P*(x)
2mJL*-
1 - 2x
n=l
Proof.
ffxlt11"1
n=l
Now let x = t,
00
11 1
= y \ (x)x - =y^p*(x2) = i.
-i
1 _ x2 _ x2
Z#
n=l
Z-#
n
n=l
_ ,
Therefore,
_op
p*(x) = - 1 = 1 + 2x + 4x2 + . + 2 n x n +
I n
1 - 2x
n=l
Notice that the polynomials P*(x)
rising diagonals of Pascal s triangle whose sums are the Fibonacci numbers but in the r e v e r s e o r d e r of those for P (x).
P(x)
Assume that
(x) = x m P * (l/x)
'
m
for m > 1 .
Pi(x) = x = x ^ P f U / x ) ]
m = 2:
P2(x) = x = x 2 (l/x) = x 2 [ P f ( l / x ) ]
m = 3:
and
324
k-l(x)
xk_lp
k_l(l/x)
P k (x) = x P * ( l / x )
[Oct.
'
.
Then, by the recurrence relations for the polynomials P (x) and P*(x) ,
k+lW
kfe)
XP
k-l(X>
= xkP*(l/x) + x x ^ P ^ U / x )
= X k + 1 ( l / x ) [ P * ( l / x ) 4- P ^ d / x ) ]
= xk+1P*+1(l/x)
Then,
an integer has a unique representation in Fibonacci numbers if and only if its minimal and
maximal representations a r e the same, which condition occurs only for the integers of the
form F
is not unique, and, from the above r e m a r k s , unless the number is one l e s s than a Fibonacci
number, it will have at least two representations in Fibonacci numbers.
stop h e r e . The Fibonacci numbers F n and F 0 ,., can each be written as the sum of d i s ^
2n
2n+l
tinct Fibonacci numbers 1, 2, 3, 5, 8, , in n different ways. F o r other integers p,
the r e a d e r is invited to experiment to see what theorems he can produce.
We now turn to representations of integers using Lucas numbers.
4.
If we change our representative set from Fibonacci numbers to Lucas numbers, we can
find minimal and maximal representations of integers as sums of distinct Lucas numbers.
The Lucas numbers are
2, 1, 3, 4, 7, 11, '* ,
defined by
L0 = 2 ,
L 4 = 1,
L 2 = 3,
L_L1=L+L
n > 1. (See Brown L[6a].)
J
n+1
n
n - 11S
The derivation of a recursion formula for the enume rating minimal polynomials Q (x)
for Lucas numbers is very s i m i l a r to that for the polynomials P (x) for Fibonacci numbers,,
Details of the proofs a r e omitted here.
< p < L
re-
-,
15 = 11
13 = 11
16 = 11 + 4
14 = 11
W = 11 + 4
197
3]
325
so that Q4(x) = 2 x + 4 x + x since 2 integers require 3 Lucas numbers, 4 integers r e quire 2 Lucas numbers, and one integer requires one Lucas number.
Notice that
L 5 = 11
is included in each representation, but that L 4 = 7 does not appear in any representation in
this inverval.
< P <
Q
^ m - 1-, (x)
'
m+1
= 1
< P < 3
= 2
< P < 4
X =
= 3
< P < 7
2x 2 + x = Q 2 (x)
= 4
< P < 11
3x 2 + x = Q 3 (x)
= 5
11
< P
<
2x = Qo(x)
2X 3 + 4x* + x =
18
= 6
18
< P < 29
= 7
29
< P < 47
= 8
47
< P < 76
= 9
76
Qi(x)
5X
+ 5x
Q 4 (x)
+ x = Q 5 (x)
2
2X 4 -* 9 x 3 + 6x + x =
Q 6 (x)
7x 4 + 14x 3 + 7x 2 + x = Q 7 (x)
2x 5 + 16x 4 + 2 Ox3 + 8x 2 + x = Qs(x)
Similarly to P (x), by the rules of polynomial addition and because of the way the polynomials Q (x) are defined,
jti
Q0(x) = 2x,
Qt(x)
= x ,
is the recursion relation satisfied by the polynomials Q (x). Here we have the same r e c u r sion formula satisfied by the polynomials P (x), but with different starting values.
that Q (1) = L . As before,
Al)
Notice
o r , the
count of the numbers for which a minimal representation exists in the interval L n < p < L n + 1
which contains exactly
- - L
= L
integers.
fill the n boxes with zeros o r ones such that no two ones a r e adjacent and to discard the
arrangement using all zeros. As before, there a r e
F n + 2 - 1 such arrangements.
Now,
LQ } L ^ li2t Lg,
into the n boxes. Again, the significance of the ones and zeros is determination
of which Lucas numbers are used in the sum. ' But, notice that
L 0 + L 2 = Lj^ + L 3 ,
which
would make m o r e than one minimal representation of an integer possible. To avoid this p r o b l e m , we consider the first four boxes and reject L 0 + L2 whenever that combination occurs
without L 4 or L 2 . If such four boxes hold then there are
(n - 4)
326
can hold F
compatible arrangements.
[Oct
+2
- F
- 1 = L
F
- 1.
(n + 1). Therefore, using the first n Lucas numbers and the two constraints, we can have
at most L - 1 different numbers represented, for
Li + L3 + L2 + L4 +
2k-1 "
2k-2
2k
2k-1
- 2
-
, but the enumerating minimal polynomial guarantees that each of the numbers 1, 2, 3,
This is the
Also, we
need to exclude the ending ht + L 3 in our representations to exclude the possibility of two
maximal representations for an integer, one using L 0 + L2 = 5 and the other
LA + L 3 = 5.
We will use the combination LA + L 3 only when one of L 0 or L2 occurs in the same maximal representation.
Now, let the enumerating maximal polynomials for the Lucas case for the interval L &
p < L
- be Q*
=
=
=
=
=
=
=
7 + 3 + 1
7 + 3 + 2
7 + 3 + 2 + 1
7 + 4 + 2 + 1
7 + 4 + 3 + 1
7 + 4 + 3 + 2
7 + 4 + 3
-f
2 + 1
so that Q|4(x) = x 5 + 4X4 + 2X3, since one integer requires 5 Lucas numbers, 4 integers r e quire 4 Lucas numbers, and 2 integers require 3 Lucas numbers in their maximal r e p r e s e n tation.
1973]
327
Q*
m+1
^m-
m = 1
1 < p < 3
2x = Qo*(x)
m = 2
3 < p < 4
x 2 = Q*(x)
m = 3
4 < p < 7
X3
+ 2x2 = Q2*(x)
m = 4
7 < p < 11
X4
+ 3x3 = Q3*(x)
m = 5
11 < p < 18
X5
+ 4x 4 + 2x3 = Q4*(x)
m = 6
18 < p < 29
X6
m = 7
29 < p < 47
m = 8
47 < p < 76
m = 9
The recursion relation for the polynomials Q*(x) can be derived in a similar fashion
to P*(x),
n
becoming
&
Q* + 1 W
Q0*(x) = 2x,
Qf(x) = x 2 .
Notice that the same coefficients occur in the enumerating minimal Lucas polynomial
Q (x)
and in the enumerating maximal Lucas polynomial Q*(x). The relationship in the lemma b e low could be proved by mathematical induction, paralleling the proof of the similar property
of P (x) and P*(x) given in the preceding section.
i
Lemma.
Q (x) = x m + 1 Q * (1/x)
^m '
^m '
for m > 1 .
nj.
= F +F
n
n *
1 ,
, when x = 1.
Q
nA
00
328
[Oct.
k-2
^
so that 32
integers require k Lucas numbers in their maximal representation, k > 2.
A proof that this is the c o r r e c t computation of the sum of the polynomials Q*(x) follows.
Lemma.
If
Qf(x) = x 2 ,
Q*(x) = 2x,
and
then
= xn+1Lfn(X)
Q * ^ )
f n _ 2 (x)] ,
n = 2:
Then
Q*(x2) = x 2 [ Q * ^ ( x 2 ) + Q*_ 2 (x 2 )]
= x 2 {x n + 1 [f n (x) + f n _ 2 (x)] + ^ [ f n - 1 W
n+2
+ f n _ 3 (x)]}
+[xfn-2W + ^ ( X ) ] }
[fn+1(x)+fn^(x)],
n=l
t2
v ,n+l
.fn (x)t
1 - xt - t2
OO
JLmJ n" 2
n+l
tfq - xt)
n=l
! - * -
Adding,
"
xt
~* 1^1
Setting t = x,
T^TT
1
Therefore,
2x
=Z)xI1+1[fn<x>
n=l
+ f
-2<x>] =^K-i^
il
1973]
,Q*(x)
2x
= 2x +
3xz
2x
3-2n-2xn
2x + >
n=0
329
n=2
L3
L2
Count of Possibilities (n ^. 4)
2n-4
excluded
2n-3
2n-3
2n"3
2n-3
2n-4
0n-3
n ^
4,
quire that the L 4 box contain a one, while all would have either an L 4 or an L 5 appearing
in the representation.
or
clude the representations of 1 through 9, 11 and 12, which give the first three t e r m s 2x + 3x2
+ 6x 3 of the enumerating maximal Lucas polynomial sum and explain the i r r e g u l a r first t e r m
in the sum of the polynomials Q*(x). The numbers not included in the count of possibilities
above follow.
J
r e p r e s en ting:
L3
L2
Li
Lo
2 \
3 J
5 '
2X
3x 2
6X3
" 1
12
330
[Oct.
k-2
Now, the enumerating maximal polynomial guarantees that 3*2 ~ integers require k
Lucas numbers in their maximal representation, but examining the possible maximal r e p r e k-2
sentations which could be written using k Lucas numbers shows that at most 3-2
different representations could be formed. That i s exactly one apiece, so the maximal r e p r e s e n tation of an integer using Lucas numbers subject to the two constraints, that no two consecutive Lucas numbers are omitted and that the combination L 3 + L 1 is not used unless L 0 or
L 2 also appear, is unique.
5. CONCLUDING REMARKS
Much interest has been shown in the subject of representations of integers in recent
years.
Some of the many diverse new results which a r i s e naturally from this paper a r e r e -
two is complete; any positive integer can be represented in the binary system of numeration.
However, if any power of 2, for example,
The Dual Zeckendorf Theorem has an extension that characterizes the Fibonacci numbers.
Brown in [2] proves that, if each positive integer has a unique representation as the sum of
distinct m e m b e r s of a given sequence when no two consecutive m e m b e r s of the sequence a r e
omitted in the representation, then the given sequence is the sequence of Fibonacci numbers.
Generalized Fibonacci numbers can be studied in a manner similar to the Lucas case.
A set of particularly interesting sequences arising in P a s c a l ' s triangle appears in [6]; the
sequences formed as the sums of elements of the diagonals of Pascal 1 s left-justified triangle,,
beginning in the left-most column and going right one and up p throughout the a r r a y .
(The
Fibonacci numbers occur when p = 1.) Or, the squares of Fibonacci numbers may be used
(see [7]), which gives a complete sequence if m e m b e r s of the sequence can be used twice.
Other ways of studying generalized Fibonacci numbers include those given in [8] , [ 9 ] , [10] ,9
and [11].
To return to the introduction, Carlitz [12] and Klarner [13] have studied the problem
of counting the number of representations possible for a given integer. Tables of the number
of representations of integers as sums of distinct elements of the Fibonacci sequence as well
as other related tables appear In [14]. The general problem of xepresentations of integers
using the Fibonacci numbers, enumerating intervals, and positional binary notation for the
representations were given by F e r n s [15] while [16] is one of the e a r l i e s t references following Daykin [8]. The suggested readings and the references given here a r e by no means exhaustive.
1973
331
REFERENCES
1. John L. Brown, J r . , "Zeckendorf's Theorem and Some A p p l i c a t i o n s / ' Fibonacci Quarterly, Vol. 2, No. 3, Oct. , 1964, pp. 163-168.
2. John L. Brown, J r . , "A New Characterization of the Fibonacci N u m b e r s , " Fibonacci
Quarterly, Vol. 3, No. 1, Feb. , 1965, pp. 1-8.
3. David A. K l a r n e r , "Representations of N as a Sum of Distinct Elements from Special
Sequences," Fibonacci Quarterly, D e c , 1966, pp. 289-306.
4.
Verner E. Hoggatt, J r . , Fibonacci and Lucas Numbers, Houghton Mifflin, Boston, 1969,
Chapter 12.
12.
13. David A. Klarner, "Partitions of N into Distinct Fibonacci Numbers," Fibonacci Quarterly, Oct. , 1968, pp. 235-244.
14.
Brother Alfred Brousseau, Fibonacci and Related Number Theoretic Tables, The Fibonacci Association, San Jose State College, San J o s e , Calif. , 1972.
Let
S(X2)
925
q.
F o r example,
= 311.
= X when X is given. F o r
=21.
Step II:
10
20
Step III:
10
20
12
15
14
Step IV:
1(20)
2(19)
4(17)
5(16)
10(11)
20(1)
3(18)
6(15)
12(9)
15(6)
7(14)
14(7)
211
106
85
43
22
141
71
36
29
61
31
157 + 4 1 1 + 1076 + 2817 + 7375 + 19308 + 50549 + 132339 + 346468 + 907065 + 2374727 = 3942292
Dividing by 199 gives 19308.
Brother Alfred Brousseau, St. Mary T s College, California
332
(a) Prove that the sum of n consecutive Lucas numbers is divisible by 5 if and only if
n is a multiple of 4.
(b) Determine the conditions under which a sum of n consecutive Lucas numbers is a
multiple of 10.
B-263
Proposed by Timothy B. Carroll, Graduate Student, Western Michigan University, Kalamazoo, Michigan.
Let S
+ 1 = 0.
(a) Find a recursion formula for S .
(b) Express S
B-264
= (-l)n+1
and
F^ - F n _ 2 F n + 2 = ( - l ) n
- 1
Let F
and L
to obtain a
334
B-266
[Oct.
Let L
B-267 Proposed by Marjorie Bicknell, A. C. Wilcox High School, Santa Clara, California.
Can you guess WHO IS SHE? This is an easy simple addition and SHE is divisible by 29.
WHO
IS
SHE
Solution by John l/l/. Milsom, Butler County Community College, Butler, Pennsylvania.
Although it is not stated in the problem, assume (as is customary) that distinct l e t t e r s
represent distinct digits.
The letter I must be replaced with the number
9 (base 10).
W is one l e s s than S.
Examining the three-digit numbers which a r e divisible by 29, there a r e three sets of numbers
which satisfy the conditions imposed by the problem.
WHO
628
714
743
IJ5
97
98
98
SHE
725
812
841
628
97
725
2.
714
98
812
3.
743
98
841
Also solved by Harold Don Allen, Paul S.Bruckman, J. A. H. Hunter, Robert Kaplar, Jr., Edgar Karst, David Zeitlin, and the
Proposer. At least one of the solutions was found by Kim Bachick, Warren Cheves, and Herta T. Freitag. A solution with W =
Ewas found by Richard W. Sielaff.
= pu n + qu ^
(n > 1).
Put
1973]
335
. i = U U
U , , - / 11- U U. ,
< !M = 1
k f
n n-1
n-k+1 ' 1 2
k'
J 01
Show that
(*)
Solution by Graham Lord, Temple University, Philadelphia, Pennsylvania.
Let
L H S kj
P - P" 2F | k - i j j k + i
When n = 2 and k = 1 the inequality is not s t r i c t since L = 0. Now
L = v(u
, ,.. u, ,. - Fp 2 u . u, )u2 u2 . l 0 u . , -/ / u2 u? u,
n-k+1 k+1
n-k k n
n-k+2 n-k+1
1
k k+1
Also
u
. , - u. , _ - F p2u . u.
n-k+1 k+1
n-k k
Since u,
n = 2, k = 1.
Also solved by H. W. Gould and the Proposer.
3F
+2
n+3
is an exact divisor
of
7F 3
- F3
- F3 .
n+2
n+1
n
Solution by Phil Mana, University of New Mexico, Albuquerque, New Mexico.
Let F
Also solved by Wray G. Brady, Paul S. Bruckman, James D. Bryant, L. Carlitz, Warren Cheves, Herta T. Freitag, J. A. H.
Hunter, Edgar Karst, Graham Lord, F. D. Parker, David Zeitlin, and the Proposer.
- F2
- - 1 and 2F 2
336
Oct. 1973
F2 + F
F
2n
2 n - l 2n+l
Since
F 0 _L1 F Q n - F 2 = 1, 2 F 0 n F 0 _,, - 1 = 2F 2 + 1 = F 2 + F 0 , F 0 x 1
2n+l 2 n - l
2n
2 n - l 2n+l
2n
2n
2 n - l 2n+l
If any one of these three equal expressions r e p r e s e n t s a p r i m e , so do the other two.
Also solved by James D. Bryant, Edgar Karst, David Zeitlin, and the Proposer.
FIBONACCI-PASCAL PROPORTION
B-242 Proposed by J. Wlodarski, Proz-Westhoven, Federal Republic of Germany.
Prove that
(W>->)
= F
- F
m m+1
for infinitely many values of the integers m, n, and k (with 0 < k < n).
Solution by the Proposer.
Let
R
Then
(0*W
/F
= (n - k + l ) / k .
.. by choosing k as tF
- and n as tF
- 1,
Prove that
( : ) * ( " )
= F
- F
m m+1
for infinitely many values of the integers m , n, and k (with 0 < k < n).
Solution by the Proposer.
Here the given ratio of binomial coefficients equals (n - k + l)/(n + 1) and this becomes
F
/F
- when n = t F
- - 1 and k = tF